<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0">
    <channel>
        <title>Cycles Blog</title>
        <link>https://runcycles.io/blog/</link>
        <description>News, guides, and updates from the Cycles team.</description>
        <lastBuildDate>Sat, 09 May 2026 16:03:32 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <language>en</language>
        <copyright>Copyright 2025-present RunCycles.io</copyright>
        <item>
            <title><![CDATA["We Put Cycles in Front of Our Own Outreach Agent"]]></title>
            <link>https://runcycles.io/blog/we-put-cycles-in-front-of-our-own-outreach-agent</link>
            <guid isPermaLink="false">https://runcycles.io/blog/we-put-cycles-in-front-of-our-own-outreach-agent</guid>
            <pubDate>Sat, 09 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A dogfood field report on governing a real autonomous outreach agent with Cycles — Gmail drafts, Slack review, budget caps, and DENY on external send."]]></description>
            <author>Evan Mavashev</author>
        </item>
        <item>
            <title><![CDATA["A Supply-Chain Playbook for Agent Skill Marketplaces"]]></title>
            <link>https://runcycles.io/blog/agent-skill-marketplace-supply-chain-playbook</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agent-skill-marketplace-supply-chain-playbook</guid>
            <pubDate>Fri, 08 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["npm and PyPI took a decade to converge on signing, OIDC publishing, and provenance. Agent skill marketplaces can skip the detour — but also need runtime limits."]]></description>
            <author>Albert Mavashev</author>
            <category>security</category>
            <category>supply-chain</category>
            <category>marketplace</category>
            <category>governance</category>
            <category>agents</category>
            <category>runtime-authority</category>
            <category>action-control</category>
            <category>production</category>
        </item>
        <item>
            <title><![CDATA["Why Local-First Agent Runtimes Need Runtime Authority"]]></title>
            <link>https://runcycles.io/blog/every-local-first-agent-runtime-needs-budget-authority</link>
            <guid isPermaLink="false">https://runcycles.io/blog/every-local-first-agent-runtime-needs-budget-authority</guid>
            <pubDate>Thu, 07 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Local-first, BYOK agent runtimes — OpenClaw, Cline, Aider, Continue — share a cost and action-risk governance gap that no provider cap or local limit closes."]]></description>
            <author>Albert Mavashev</author>
            <category>agents</category>
            <category>governance</category>
            <category>runtime-authority</category>
            <category>local-first</category>
            <category>byok</category>
            <category>budgets</category>
            <category>cost-control</category>
            <category>comparisons</category>
            <category>best-practices</category>
            <category>marketplace</category>
        </item>
        <item>
            <title><![CDATA["Policy Drift in AI Agents"]]></title>
            <link>https://runcycles.io/blog/policy-drift-in-ai-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/policy-drift-in-ai-agents</guid>
            <pubDate>Thu, 07 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["AI agent policies drift when prompts, skills, tools, models, and workflows change. Static approval needs shadow mode, runtime limits, and audit loops."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["AI Agent Kill Switches Should Be Scoped"]]></title>
            <link>https://runcycles.io/blog/ai-agent-kill-switches-should-be-scoped</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-kill-switches-should-be-scoped</guid>
            <pubDate>Wed, 06 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Global AI agent kill switches are blunt incident tools. Scoped freezes, suspends, pauses, and budget limits contain damage without stopping everything."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Your First Week with Cycles Budget Guard for OpenClaw"]]></title>
            <link>https://runcycles.io/blog/openclaw-budget-guard-first-week-dry-run-to-production</link>
            <guid isPermaLink="false">https://runcycles.io/blog/openclaw-budget-guard-first-week-dry-run-to-production</guid>
            <pubDate>Wed, 06 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["An operator playbook for the first week after installing cycles-openclaw-budget-guard: simulated dry-run, calibration, then a canary cutover decision."]]></description>
            <author>Albert Mavashev</author>
            <category>openclaw</category>
            <category>budgets</category>
            <category>agents</category>
            <category>production</category>
            <category>operations</category>
            <category>dry-run</category>
            <category>cost-control</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["Python AI Agent Control: Cost, Risk, and Audit by Layer"]]></title>
            <link>https://runcycles.io/blog/python-ai-agent-control-cost-risk-audit-layers</link>
            <guid isPermaLink="false">https://runcycles.io/blog/python-ai-agent-control-cost-risk-audit-layers</guid>
            <pubDate>Wed, 06 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Python AI agent tools cover cost OK, risk partially, and audit barely. Six layers of agent control, what each actually does, and where each stops short."]]></description>
            <author>Albert Mavashev</author>
            <category>agents</category>
            <category>governance</category>
            <category>runtime-authority</category>
            <category>action-control</category>
            <category>audit</category>
            <category>costs</category>
            <category>risk-assessment</category>
            <category>python</category>
            <category>production</category>
        </item>
        <item>
            <title><![CDATA["Agent Registries Are Not Runtime Governance"]]></title>
            <link>https://runcycles.io/blog/agent-registries-are-not-runtime-governance</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agent-registries-are-not-runtime-governance</guid>
            <pubDate>Tue, 05 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Agent registries help inventory owners, lifecycle, and access, but runtime governance still needs per-action budget, risk, scope, and audit decisions."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["How scalerX.ai Wired Cycles Into a Java Agent Runtime"]]></title>
            <link>https://runcycles.io/blog/how-scalerx-wired-cycles-into-a-java-agent-runtime</link>
            <guid isPermaLink="false">https://runcycles.io/blog/how-scalerx-wired-cycles-into-a-java-agent-runtime</guid>
            <pubDate>Tue, 05 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How scalerX.ai added budget reservations to a Java Spring Boot agent runtime — the @Cycles annotation, scope hierarchies, and reserve/commit on OpenAI calls."]]></description>
            <author>Cycles Team</author>
        </item>
        <item>
            <title><![CDATA["Add Hard Budgets to MCP Tools Before They Execute"]]></title>
            <link>https://runcycles.io/blog/mcp-tool-budgets-before-execution</link>
            <guid isPermaLink="false">https://runcycles.io/blog/mcp-tool-budgets-before-execution</guid>
            <pubDate>Tue, 05 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["MCP exposes tools, but it does not decide whether the next call should run. Wrap MCP handlers with Cycles reserve/commit checks before side effects happen."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Agent Skills Are the New Supply Chain"]]></title>
            <link>https://runcycles.io/blog/agent-skills-are-the-new-supply-chain</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agent-skills-are-the-new-supply-chain</guid>
            <pubDate>Mon, 04 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Agent skills turn reusable workflows into executable supply chain risk. Govern them with inventory, provenance, sandboxing, runtime limits, and audit."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["MCP Gateways Are Not Runtime Authority"]]></title>
            <link>https://runcycles.io/blog/mcp-gateways-are-not-runtime-authority</link>
            <guid isPermaLink="false">https://runcycles.io/blog/mcp-gateways-are-not-runtime-authority</guid>
            <pubDate>Sun, 03 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["MCP gateways help secure tool connectivity, but production agents still need runtime authority for budget, risk, scope, and per-action decisions at runtime."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Agent Identity Is Not User Identity"]]></title>
            <link>https://runcycles.io/blog/agent-identity-is-not-user-identity</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agent-identity-is-not-user-identity</guid>
            <pubDate>Sat, 02 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Why production AI agents need dedicated identities, scoped credentials, owner mapping, audit trails, and runtime authority instead of borrowed user sessions."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Webhook Delivery That Operators Can Trust"]]></title>
            <link>https://runcycles.io/blog/webhook-delivery-operators-can-trust</link>
            <guid isPermaLink="false">https://runcycles.io/blog/webhook-delivery-operators-can-trust</guid>
            <pubDate>Fri, 01 May 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A production webhook delivery contract for AI agent events: signed bodies, retries, dedupe keys, stale cutoffs, auto-disable, replay, and traceability."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Bulk Operations Are Incident Response Infrastructure"]]></title>
            <link>https://runcycles.io/blog/bulk-operations-ai-agent-incident-response</link>
            <guid isPermaLink="false">https://runcycles.io/blog/bulk-operations-ai-agent-incident-response</guid>
            <pubDate>Thu, 30 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Why AI agent platforms need previewed, idempotent bulk actions to contain tenant, webhook, and budget incidents without blind production scripts at 2 AM."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Least-Privilege API Keys for AI Agents"]]></title>
            <link>https://runcycles.io/blog/least-privilege-api-keys-for-ai-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/least-privilege-api-keys-for-ai-agents</guid>
            <pubDate>Wed, 29 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How to scope AI agent API keys by tenant, environment, and permission so one leaked credential cannot become cross-tenant budget authority in production."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Cursor Agent Deleted a Prod Database in 9 Seconds"]]></title>
            <link>https://runcycles.io/blog/ai-agent-deleted-prod-database-9-seconds</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-deleted-prod-database-9-seconds</guid>
            <pubDate>Tue, 28 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Cursor agent reportedly wiped a Railway production database and backups in one API call. Prompts aren't permissions — agents need pre-execution gates."]]></description>
            <author>Albert Mavashev</author>
            <category>incidents</category>
            <category>runtime-authority</category>
            <category>governance</category>
            <category>audit</category>
            <category>action-authority</category>
            <category>agents</category>
        </item>
        <item>
            <title><![CDATA["The AI Agent Audit Trail You're Already Building"]]></title>
            <link>https://runcycles.io/blog/runtime-authority-byproducts-audit-trail-and-attribution-by-default</link>
            <guid isPermaLink="false">https://runcycles.io/blog/runtime-authority-byproducts-audit-trail-and-attribution-by-default</guid>
            <pubDate>Mon, 27 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Runtime authority's hidden byproduct: ledger-ready audit, cost, and attribution records for AI agents — useful for compliance, FinOps, and platform chargeback."]]></description>
            <author>Cycles Team</author>
            <category>runtime-authority</category>
            <category>governance</category>
            <category>audit</category>
            <category>finops</category>
            <category>compliance</category>
            <category>costs</category>
            <category>agents</category>
            <category>engineering</category>
        </item>
        <item>
            <title><![CDATA["Why Multi-Agent Coordination Fails — and What Actually Prevents It"]]></title>
            <link>https://runcycles.io/blog/multi-agent-coordination-failure-structural-prevention</link>
            <guid isPermaLink="false">https://runcycles.io/blog/multi-agent-coordination-failure-structural-prevention</guid>
            <pubDate>Sun, 26 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Multi-agent failures aren't a prompt problem. They're structural. A look at the UC Berkeley MAST taxonomy and the prevention patterns that actually work."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Deploying the Cycles Events Service at Scale"]]></title>
            <link>https://runcycles.io/blog/deploying-cycles-events-service-at-scale</link>
            <guid isPermaLink="false">https://runcycles.io/blog/deploying-cycles-events-service-at-scale</guid>
            <pubDate>Sat, 25 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How to run the Cycles events service in production: the 7980/9980 port split, Prometheus scraping, Kubernetes probes, and alert rules that actually page the right people."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Webhook Idempotency Patterns for AI Agent Budget Events"]]></title>
            <link>https://runcycles.io/blog/webhook-idempotency-patterns-for-ai-agent-budget-events</link>
            <guid isPermaLink="false">https://runcycles.io/blog/webhook-idempotency-patterns-for-ai-agent-budget-events</guid>
            <pubDate>Fri, 24 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How to design idempotent webhook receivers for Cycles budget events: X-Cycles-Event-Id dedup, HMAC verification, Redis vs Postgres patterns, and replay testing."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Shadow Mode to Hard Enforcement: The Cutover Decision Tree"]]></title>
            <link>https://runcycles.io/blog/shadow-to-enforcement-cutover-decision-tree</link>
            <guid isPermaLink="false">https://runcycles.io/blog/shadow-to-enforcement-cutover-decision-tree</guid>
            <pubDate>Thu, 23 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["When is an AI agent budget policy actually ready for hard enforcement? A signal-driven decision tree — not a calendar — for flipping from dry-run to blocking."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Tenant Lifecycle at Scale: Cascade Semantics"]]></title>
            <link>https://runcycles.io/blog/tenant-lifecycle-cascade-semantics-at-scale</link>
            <guid isPermaLink="false">https://runcycles.io/blog/tenant-lifecycle-cascade-semantics-at-scale</guid>
            <pubDate>Thu, 23 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["What happens when you close an AI-agent tenant matters more than how you close it. A walk through cascade semantics, terminal-owner guards, and the zombie-budget problem."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["W3C Trace Context for AI Agent Debugging"]]></title>
            <link>https://runcycles.io/blog/w3c-trace-context-ai-agent-debugging</link>
            <guid isPermaLink="false">https://runcycles.io/blog/w3c-trace-context-ai-agent-debugging</guid>
            <pubDate>Thu, 23 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How Cycles wires W3C Trace Context across admin, runtime, events, and audit planes so you can correlate an AI agent's budget decisions across every plane at once."]]></description>
            <author>Albert Mavashev</author>
        </item>
        <item>
            <title><![CDATA["Budgeting Reasoning Tokens: Governing Extended Thinking"]]></title>
            <link>https://runcycles.io/blog/budgeting-reasoning-tokens-governing-extended-thinking-before-it-bills</link>
            <guid isPermaLink="false">https://runcycles.io/blog/budgeting-reasoning-tokens-governing-extended-thinking-before-it-bills</guid>
            <pubDate>Wed, 22 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Reasoning tokens are billed but often hidden from callers. Output-length heuristics and provider caps don't govern them cleanly. Here's how to cap at runtime."]]></description>
            <author>Albert Mavashev</author>
            <category>governance</category>
            <category>budgets</category>
            <category>production</category>
            <category>costs</category>
            <category>risk</category>
        </item>
        <item>
            <title><![CDATA["Tracking Tokens and Cost in a Streaming LLM Response"]]></title>
            <link>https://runcycles.io/blog/tracking-tokens-in-a-streaming-llm-response</link>
            <guid isPermaLink="false">https://runcycles.io/blog/tracking-tokens-in-a-streaming-llm-response</guid>
            <pubDate>Sun, 19 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Streaming LLM cost tracking breaks reserve-commit in four ways. A Python context manager for OpenAI + Anthropic budget enforcement that handles all four."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>streaming</category>
            <category>llm</category>
            <category>budget-enforcement</category>
            <category>python</category>
            <category>openai</category>
        </item>
        <item>
            <title><![CDATA["How We Stopped Shipping Performance Regressions"]]></title>
            <link>https://runcycles.io/blog/how-we-stopped-shipping-performance-regressions</link>
            <guid isPermaLink="false">https://runcycles.io/blog/how-we-stopped-shipping-performance-regressions</guid>
            <pubDate>Sat, 18 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A two-tier CI benchmark gate for the Cycles release process: release-blocking at 25% vs a pinned baseline, trend-warning at 30% vs a rolling 7-run median, with 3-trial medians to absorb GitHub-runner noise."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>operations</category>
            <category>performance</category>
            <category>ci</category>
            <category>benchmarking</category>
            <category>regression-testing</category>
        </item>
        <item>
            <title><![CDATA["Designing a Redis TTL Sweeper That Doesn't Lie"]]></title>
            <link>https://runcycles.io/blog/designing-a-redis-ttl-sweeper-that-doesnt-lie</link>
            <guid isPermaLink="false">https://runcycles.io/blog/designing-a-redis-ttl-sweeper-that-doesnt-lie</guid>
            <pubDate>Fri, 17 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A TTL sweeper sits on top of your ledger — it must never double-act, miss a candidate, or lose itself to clock skew. Here are seven properties a correct one enforces, and two bugs we shipped before we got there."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>architecture</category>
            <category>operations</category>
            <category>redis</category>
            <category>lua</category>
            <category>background-jobs</category>
            <category>ttl</category>
        </item>
        <item>
            <title><![CDATA["Where Did My Tokens Go? Debugging Agent Spend at Production Scale"]]></title>
            <link>https://runcycles.io/blog/where-did-my-tokens-go-debugging-agent-spend</link>
            <guid isPermaLink="false">https://runcycles.io/blog/where-did-my-tokens-go-debugging-agent-spend</guid>
            <pubDate>Fri, 17 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Debug AI agent spend with scope paths, event streams, and correlation IDs — what Cycles' spend-debugging events and balance API actually let you pivot on when the bill triples."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>debugging</category>
            <category>observability</category>
            <category>agents</category>
            <category>cost-attribution</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["Why We Added Property Tests to Cycles' Budget Authority"]]></title>
            <link>https://runcycles.io/blog/why-we-added-property-tests-to-cycles-budget-authority</link>
            <guid isPermaLink="false">https://runcycles.io/blog/why-we-added-property-tests-to-cycles-budget-authority</guid>
            <pubDate>Thu, 16 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Ordinary unit tests won't systematically explore the interleavings that break atomicity. Here's the property test we built for Cycles' budget authority — and two jqwik-Spring traps worth knowing."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>testing</category>
            <category>operations</category>
            <category>property-based-testing</category>
            <category>jqwik</category>
            <category>concurrency</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["Why Cycles Runs Budget Authority on Redis Lua"]]></title>
            <link>https://runcycles.io/blog/why-cycles-runs-budget-authority-on-redis-lua</link>
            <guid isPermaLink="false">https://runcycles.io/blog/why-cycles-runs-budget-authority-on-redis-lua</guid>
            <pubDate>Wed, 15 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Budget authority sits in the hot path of every agent action. Three constraints — atomicity, sub-10ms envelope, correctness under retry storms — named the substrate. Here's what we picked and what we gave up."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>architecture</category>
            <category>operations</category>
            <category>redis</category>
            <category>lua</category>
            <category>concurrency</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["Agents Are Cross-Cutting. Your Controls Aren't."]]></title>
            <link>https://runcycles.io/blog/agents-are-cross-cutting-your-controls-arent</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agents-are-cross-cutting-your-controls-arent</guid>
            <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Agents span providers, tools, tenants, and workers. Tool-local controls govern only one slice — here is why agent governance has to be cross-cutting."]]></description>
            <author>Albert Mavashev</author>
            <category>runtime-authority</category>
            <category>governance</category>
            <category>architecture</category>
            <category>comparisons</category>
            <category>multi-tenant</category>
            <category>multi-agent</category>
            <category>observability</category>
        </item>
        <item>
            <title><![CDATA["The 429 Classification Gap: Why Your Retry Logic Is Lying to You"]]></title>
            <link>https://runcycles.io/blog/429-classification-gap-why-retry-logic-lying</link>
            <guid isPermaLink="false">https://runcycles.io/blog/429-classification-gap-why-retry-logic-lying</guid>
            <pubDate>Fri, 10 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["HTTP 429 is not one failure — it is three: WAIT, CAP, STOP. Why classifying 429s before retry is the missing primitive in AI agent runtimes."]]></description>
            <author>Brenton Williams</author>
            <category>ai-agents</category>
            <category>retry-logic</category>
            <category>retry-storms</category>
            <category>rate-limits</category>
            <category>http-429</category>
            <category>429</category>
            <category>agent-infrastructure</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Governance Dashboard: Budgets, Risk, Keys"]]></title>
            <link>https://runcycles.io/blog/ai-agent-governance-admin-dashboard-monitor-control-budgets-risk</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-governance-admin-dashboard-monitor-control-budgets-risk</guid>
            <pubDate>Thu, 09 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["AI agents have budgets and risk limits. But when something breaks at 2am, what do you actually do? Three production scenarios, the pattern behind governance UIs, and why read-only dashboards aren't enough."]]></description>
            <author>Albert Mavashev</author>
            <category>product</category>
            <category>operations</category>
            <category>dashboard</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["Retry Storms & Idempotency in Agent Budget Systems"]]></title>
            <link>https://runcycles.io/blog/retry-storms-and-idempotency-in-agent-budget-systems</link>
            <guid isPermaLink="false">https://runcycles.io/blog/retry-storms-and-idempotency-in-agent-budget-systems</guid>
            <pubDate>Wed, 08 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["AI agent retries amplify both cost and risk — same email sent twice, same budget charged twice. Idempotency keys are the primitive that prevents it."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>production</category>
            <category>operations</category>
            <category>best-practices</category>
            <category>runtime-authority</category>
            <category>architecture</category>
        </item>
        <item>
            <title><![CDATA["State of AI Agent Governance 2026"]]></title>
            <link>https://runcycles.io/blog/state-of-ai-agent-governance-2026</link>
            <guid isPermaLink="false">https://runcycles.io/blog/state-of-ai-agent-governance-2026</guid>
            <pubDate>Wed, 08 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["AI agents moved to production faster than governance kept up. This is the state of enforcement, regulation, and incidents in 2026 — and what comes next."]]></description>
            <author>Albert Mavashev</author>
            <category>governance</category>
            <category>production</category>
            <category>operations</category>
            <category>runtime-authority</category>
            <category>agents</category>
            <category>security</category>
            <category>best-practices</category>
            <category>compliance</category>
            <category>regulation</category>
            <category>incidents</category>
        </item>
        <item>
            <title><![CDATA["Estimate Drift: The Silent Killer of Budget Enforcement"]]></title>
            <link>https://runcycles.io/blog/estimate-drift-silent-killer-of-enforcement</link>
            <guid isPermaLink="false">https://runcycles.io/blog/estimate-drift-silent-killer-of-enforcement</guid>
            <pubDate>Tue, 07 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Cost estimates drift in AI agent production. When reserve:commit ratios wander outside 0.8-1.2, budgets lie. Detect drift early and recalibrate safely."]]></description>
            <author>Albert Mavashev</author>
            <category>operations</category>
            <category>production</category>
            <category>observability</category>
            <category>runtime-authority</category>
            <category>incident-response</category>
            <category>calibration</category>
        </item>
        <item>
            <title><![CDATA["Operational Runbook: Using Cycles Runtime Events"]]></title>
            <link>https://runcycles.io/blog/operational-runbook-using-cycles-runtime-events</link>
            <guid isPermaLink="false">https://runcycles.io/blog/operational-runbook-using-cycles-runtime-events</guid>
            <pubDate>Tue, 07 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Wire Cycles runtime events into on-call: severity tiers, diagnostic trees, and auto-remediation patterns for critical AI agent budget events."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>operations</category>
            <category>production</category>
            <category>incident-response</category>
            <category>webhooks</category>
            <category>observability</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Cost Control in 2026: A Landscape Guide"]]></title>
            <link>https://runcycles.io/blog/ai-agent-cost-control-2026-litellm-helicone-openrouter-runtime-authority</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-cost-control-2026-litellm-helicone-openrouter-runtime-authority</guid>
            <pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["LiteLLM, Helicone, and OpenRouter each solve part of agent cost control. What each does, where they stop, and the layer none of them cover."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>production</category>
            <category>costs</category>
            <category>agents</category>
            <category>best-practices</category>
            <category>governance</category>
            <category>architecture</category>
        </item>
        <item>
            <title><![CDATA["How to Add Runtime Enforcement Without Breaking Your Agents"]]></title>
            <link>https://runcycles.io/blog/how-to-add-runtime-enforcement-without-breaking-your-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/how-to-add-runtime-enforcement-without-breaking-your-agents</guid>
            <pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Shadow mode is the safe way to add runtime enforcement to AI agents. Roll out observe-first, calibrate, then enforce without breaking production."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>production</category>
            <category>best-practices</category>
            <category>agents</category>
            <category>runtime-authority</category>
            <category>architecture</category>
            <category>shadow-mode</category>
        </item>
        <item>
            <title><![CDATA["When Budget Runs Out: AI Agent Degradation Patterns"]]></title>
            <link>https://runcycles.io/blog/when-budget-runs-out-graceful-degradation-patterns-for-ai-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/when-budget-runs-out-graceful-degradation-patterns-for-ai-agents</guid>
            <pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Your guardrails block an agent action. Now what? Five graceful degradation patterns for handling DENY and ALLOW_WITH_CAPS — from model fallback to inform-and-stop."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>best-practices</category>
            <category>agents</category>
            <category>production</category>
            <category>action-control</category>
            <category>costs</category>
            <category>risk</category>
            <category>budgets</category>
        </item>
        <item>
            <title><![CDATA["How Teams Control AI Agents Today — And Where It Breaks"]]></title>
            <link>https://runcycles.io/blog/how-teams-control-ai-agents-today-and-where-it-breaks</link>
            <guid isPermaLink="false">https://runcycles.io/blog/how-teams-control-ai-agents-today-and-where-it-breaks</guid>
            <pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Most teams control AI agents with system prompts, rate limits, or dashboards. Here's why each breaks — and why risk, not just cost, is the real gap."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>risk</category>
            <category>governance</category>
            <category>agents</category>
            <category>best-practices</category>
            <category>production</category>
            <category>costs</category>
            <category>security</category>
            <category>multi-agent</category>
            <category>action-control</category>
        </item>
        <item>
            <title><![CDATA["We Built a Custom Agent Rate Limiter. Here's Why We Stopped."]]></title>
            <link>https://runcycles.io/blog/we-built-a-custom-agent-rate-limiter-heres-why-we-stopped</link>
            <guid isPermaLink="false">https://runcycles.io/blog/we-built-a-custom-agent-rate-limiter-heres-why-we-stopped</guid>
            <pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A post-mortem of the multi-provider AI agent rate limiter we built at scalerX — three versions, three walls, and why runtime authority is the real problem."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>production</category>
            <category>costs</category>
            <category>agents</category>
            <category>best-practices</category>
            <category>budgets</category>
            <category>runtime-authority</category>
            <category>architecture</category>
        </item>
        <item>
            <title><![CDATA["Agent Delegation Chains Need Authority Attenuation, Not Trust Propagation"]]></title>
            <link>https://runcycles.io/blog/agent-delegation-chains-authority-attenuation-not-trust-propagation</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agent-delegation-chains-authority-attenuation-not-trust-propagation</guid>
            <pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Multi-agent delegation chains propagate full trust by default. Every hop should narrow budget, actions, and scope. Here's the authority attenuation pattern."]]></description>
            <author>Albert Mavashev</author>
            <category>delegation</category>
            <category>multi-agent</category>
            <category>governance</category>
            <category>runtime-authority</category>
            <category>zero-trust</category>
            <category>security</category>
            <category>architecture</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Budget Enforcement Latency: Cycles Server Performance Benchmarks"]]></title>
            <link>https://runcycles.io/blog/cycles-server-performance-benchmarks</link>
            <guid isPermaLink="false">https://runcycles.io/blog/cycles-server-performance-benchmarks</guid>
            <pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How much latency does AI agent budget enforcement add? Published p50/p95/p99 benchmarks for every Cycles operation — 2,870+ ops/sec at 32 threads, zero errors across five versions."]]></description>
            <author>Cycles Team</author>
            <category>engineering</category>
            <category>performance</category>
            <category>benchmarks</category>
            <category>scaling</category>
            <category>latency</category>
            <category>throughput</category>
            <category>redis</category>
        </item>
        <item>
            <title><![CDATA["Why I'm Building Cycles"]]></title>
            <link>https://runcycles.io/blog/why-i-am-building-cycles</link>
            <guid isPermaLink="false">https://runcycles.io/blog/why-i-am-building-cycles</guid>
            <pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["After nearly three decades building middleware governance, I saw the same catastrophic gap emerge in AI agents. This is why I'm building Cycles."]]></description>
            <author>Albert Mavashev</author>
            <category>founder</category>
            <category>vision</category>
            <category>agents</category>
            <category>governance</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Risk Assessment: Score, Classify, Enforce"]]></title>
            <link>https://runcycles.io/blog/ai-agent-risk-assessment-score-classify-enforce-tool-risk</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-risk-assessment-score-classify-enforce-tool-risk</guid>
            <pubDate>Fri, 03 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Practical guide to AI agent risk: classify actions by blast radius, assign risk scores, and convert assessments into enforceable runtime budgets."]]></description>
            <author>Albert Mavashev</author>
            <category>risk-assessment</category>
            <category>risk-scoring</category>
            <category>action-control</category>
            <category>RISK_POINTS</category>
            <category>agents</category>
            <category>production</category>
            <category>governance</category>
        </item>
        <item>
            <title><![CDATA["The State of AI Agent Incidents (2026)"]]></title>
            <link>https://runcycles.io/blog/state-of-ai-agent-incidents-2026</link>
            <guid isPermaLink="false">https://runcycles.io/blog/state-of-ai-agent-incidents-2026</guid>
            <pubDate>Fri, 03 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Documented AI agent incidents and failure patterns — runaway costs, action misfires, security exploits, multi-agent cascades — scored by cost and blast radius."]]></description>
            <author>Albert Mavashev</author>
            <category>incidents</category>
            <category>governance</category>
            <category>security</category>
            <category>costs</category>
            <category>agents</category>
            <category>production</category>
            <category>MCP</category>
            <category>OWASP</category>
            <category>multi-agent</category>
        </item>
        <item>
            <title><![CDATA["26 Integrations: Every AI Framework, One Budget Protocol"]]></title>
            <link>https://runcycles.io/blog/26-integrations-every-ai-framework-one-budget-protocol</link>
            <guid isPermaLink="false">https://runcycles.io/blog/26-integrations-every-ai-framework-one-budget-protocol</guid>
            <pubDate>Thu, 02 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Cycles now ships 26 integrations across Python, TypeScript, Java, and Rust. One protocol enforces spend limits, action boundaries, and risk controls across every agent stack — before execution."]]></description>
            <author>Albert Mavashev</author>
            <category>announcement</category>
            <category>integrations</category>
            <category>langchain</category>
            <category>langgraph</category>
            <category>autogen</category>
            <category>openai</category>
            <category>anthropic</category>
            <category>groq</category>
            <category>django</category>
            <category>nextjs</category>
            <category>flask</category>
            <category>anyagent</category>
            <category>runtime-authority</category>
        </item>
        <item>
            <title><![CDATA["AI Agents: EU AI Act, NIST, ISO 42001, OWASP Map"]]></title>
            <link>https://runcycles.io/blog/ai-agent-governance-framework-nist-eu-ai-act-iso-42001-owasp-runtime-enforcement</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-governance-framework-nist-eu-ai-act-iso-42001-owasp-runtime-enforcement</guid>
            <pubDate>Thu, 02 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Map EU AI Act, NIST AI RMF, ISO 42001, and OWASP regulatory requirements to runtime enforcement controls — a practical governance framework for AI agents."]]></description>
            <author>Cycles Team</author>
            <category>governance</category>
            <category>compliance</category>
            <category>EU AI Act</category>
            <category>NIST</category>
            <category>ISO 42001</category>
            <category>OWASP</category>
            <category>runtime authority</category>
            <category>agents</category>
        </item>
        <item>
            <title><![CDATA["Beyond Budget: How Cycles Controls Agent Actions, Not Just Spend"]]></title>
            <link>https://runcycles.io/blog/beyond-budget-how-cycles-controls-agent-actions</link>
            <guid isPermaLink="false">https://runcycles.io/blog/beyond-budget-how-cycles-controls-agent-actions</guid>
            <pubDate>Thu, 02 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Cycles isn't just a budget tool. It governs what agents do — tool calls, API requests, emails, deployments — using the same reserve-commit protocol that controls spend. Here's how action authority works."]]></description>
            <author>Albert Mavashev</author>
            <category>action-authority</category>
            <category>risk-points</category>
            <category>runtime-authority</category>
            <category>tool-governance</category>
            <category>agents</category>
        </item>
        <item>
            <title><![CDATA["When Budget Enforcement Fires: An Operator's Guide to Cycles in Production"]]></title>
            <link>https://runcycles.io/blog/operating-budget-enforcement-in-production</link>
            <guid isPermaLink="false">https://runcycles.io/blog/operating-budget-enforcement-in-production</guid>
            <pubDate>Wed, 01 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["What to do when reservation.denied fires at 2am. Diagnostic decision trees, emergency playbooks, and metrics that predict budget incidents."]]></description>
            <author>Albert Mavashev</author>
            <category>operations</category>
            <category>incident-response</category>
            <category>production</category>
            <category>observability</category>
        </item>
        <item>
            <title><![CDATA["Real-Time Budget Alerts for AI Agents: Designing Cycles' Webhook Event System"]]></title>
            <link>https://runcycles.io/blog/real-time-budget-alerts-for-ai-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/real-time-budget-alerts-for-ai-agents</guid>
            <pubDate>Wed, 01 Apr 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["How we designed a webhook event system that delivers AI agent budget alerts to PagerDuty, Slack, and custom systems — architecture, delivery guarantees, and failure handling."]]></description>
            <author>Albert Mavashev</author>
            <category>engineering</category>
            <category>webhooks</category>
            <category>architecture</category>
            <category>observability</category>
        </item>
        <item>
            <title><![CDATA["How to Add Budget and Action Guardrails to Rust AI Agents with Cycles"]]></title>
            <link>https://runcycles.io/blog/how-to-add-budget-and-action-guardrails-to-rust-ai-agents-with-cycles</link>
            <guid isPermaLink="false">https://runcycles.io/blog/how-to-add-budget-and-action-guardrails-to-rust-ai-agents-with-cycles</guid>
            <pubDate>Tue, 31 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Add budget, action, and audit authority to Rust AI agents — control spend, tool access, token limits, step counts, and produce signed audit events for compliance and incident review, with compile-time safety."]]></description>
            <author>Albert Mavashev</author>
            <category>rust</category>
            <category>agents</category>
            <category>engineering</category>
            <category>costs</category>
            <category>governance</category>
            <category>audit</category>
            <category>guide</category>
        </item>
        <item>
            <title><![CDATA["5 AI Agent Failures Only Action Controls Would Prevent"]]></title>
            <link>https://runcycles.io/blog/ai-agent-action-failures-runtime-authority-prevents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-action-failures-runtime-authority-prevents</guid>
            <pubDate>Mon, 30 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Five AI agent failures where model spend was under $5 but the business impact was severe — and how action authority with risk-point budgets prevents each one."]]></description>
            <author>Cycles Team</author>
            <category>action-control</category>
            <category>risk</category>
            <category>incidents</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["OpenAI Agents SDK: Content Guardrails, No Action Control"]]></title>
            <link>https://runcycles.io/blog/openai-agents-sdk-has-guardrails-for-content-but-nothing-for-actions</link>
            <guid isPermaLink="false">https://runcycles.io/blog/openai-agents-sdk-has-guardrails-for-content-but-nothing-for-actions</guid>
            <pubDate>Mon, 30 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["OpenAI Agents SDK tool guardrails validate individual function tools. They aren't cross-tenant budget or risk authority — RunHooks is where that fits."]]></description>
            <author>Albert Mavashev</author>
            <category>openai</category>
            <category>agents</category>
            <category>runtime-authority</category>
            <category>governance</category>
            <category>risk</category>
            <category>actions</category>
            <category>python</category>
            <category>RunHooks</category>
        </item>
        <item>
            <title><![CDATA["Why AI Agent Tests Pass but Production Fails"]]></title>
            <link>https://runcycles.io/blog/why-ai-agent-tests-pass-but-production-keeps-failing</link>
            <guid isPermaLink="false">https://runcycles.io/blog/why-ai-agent-tests-pass-but-production-keeps-failing</guid>
            <pubDate>Mon, 30 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Only 52% of teams run agent evals, yet 89% have observability. The gap between passing tests and production is architectural — here's what closes it."]]></description>
            <author>Cycles Team</author>
            <category>agents</category>
            <category>testing</category>
            <category>evaluation</category>
            <category>production</category>
            <category>reliability</category>
            <category>CI-CD</category>
            <category>engineering</category>
        </item>
        <item>
            <title><![CDATA["Multi-Agent Systems Fail 87% — What It Costs"]]></title>
            <link>https://runcycles.io/blog/why-multi-agent-systems-fail-87-percent-cost-of-every-coordination-breakdown</link>
            <guid isPermaLink="false">https://runcycles.io/blog/why-multi-agent-systems-fail-87-percent-cost-of-every-coordination-breakdown</guid>
            <pubDate>Sun, 29 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["UC Berkeley's MAST taxonomy found 14 failure modes across 1,600+ multi-agent traces with 41–87% failure rates. We built the cost model."]]></description>
            <author>Cycles Team</author>
            <category>multi-agent</category>
            <category>failures</category>
            <category>cost</category>
            <category>coordination</category>
            <category>production</category>
            <category>MAST</category>
            <category>runtime-authority</category>
            <category>engineering</category>
        </item>
        <item>
            <title><![CDATA["We Gave Our OpenClaw Agent a $5 Budget"]]></title>
            <link>https://runcycles.io/blog/openclaw-budget-guard-five-dollar-agent</link>
            <guid isPermaLink="false">https://runcycles.io/blog/openclaw-budget-guard-five-dollar-agent</guid>
            <pubDate>Sat, 28 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A $12 OpenClaw research session constrained to a $5 Cycles budget. The agent downgrades models, disables expensive tools, self-regulates, finishes at $4.85."]]></description>
            <author>Albert Mavashev</author>
            <category>openclaw</category>
            <category>budgets</category>
            <category>agents</category>
            <category>graceful-degradation</category>
            <category>model-downgrade</category>
            <category>production</category>
            <category>cost-control</category>
            <category>ai-agent-cost</category>
            <category>llm-cost-management</category>
        </item>
        <item>
            <title><![CDATA["Five Lessons from Building a Production OpenClaw Plugin"]]></title>
            <link>https://runcycles.io/blog/openclaw-plugin-lessons-learned</link>
            <guid isPermaLink="false">https://runcycles.io/blog/openclaw-plugin-lessons-learned</guid>
            <pubDate>Sat, 28 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["We built a budget enforcement plugin for OpenClaw and hit five undocumented behaviors — including the discovery that you can't actually block a model call. Here are the workarounds we shipped and the feature requests we filed."]]></description>
            <author>Albert Mavashev</author>
            <category>openclaw</category>
            <category>plugins</category>
            <category>engineering</category>
            <category>hooks</category>
            <category>workarounds</category>
            <category>developer-experience</category>
            <category>production</category>
            <category>openclaw-plugin-development</category>
        </item>
        <item>
            <title><![CDATA["Salesforce, ServiceNow, and the Cross-Platform Gap"]]></title>
            <link>https://runcycles.io/blog/cross-platform-ai-agent-governance-salesforce-servicenow</link>
            <guid isPermaLink="false">https://runcycles.io/blog/cross-platform-ai-agent-governance-salesforce-servicenow</guid>
            <pubDate>Fri, 27 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Salesforce and ServiceNow have invested in AI agent governance. Neither acts as a neutral, shared pre-execution control plane across both — and custom runtimes."]]></description>
            <author>Albert Mavashev</author>
            <category>governance</category>
            <category>enterprise</category>
            <category>salesforce</category>
            <category>servicenow</category>
            <category>agentforce</category>
            <category>cross-platform</category>
            <category>action-authority</category>
        </item>
        <item>
            <title><![CDATA["MCP Tool Poisoning: 84% Success Rate"]]></title>
            <link>https://runcycles.io/blog/mcp-tool-poisoning-why-agent-frameworks-cant-prevent-it</link>
            <guid isPermaLink="false">https://runcycles.io/blog/mcp-tool-poisoning-why-agent-frameworks-cant-prevent-it</guid>
            <pubDate>Fri, 27 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["In benchmarks, tool poisoning attacks succeed 84% of the time with auto-approval. 10,000+ MCP servers, 30+ CVEs — and no external enforcement layer."]]></description>
            <author>Albert Mavashev</author>
            <category>security</category>
            <category>MCP</category>
            <category>tool-poisoning</category>
            <category>agents</category>
            <category>production</category>
            <category>OWASP</category>
            <category>runtime-authority</category>
            <category>supply-chain</category>
        </item>
        <item>
            <title><![CDATA["Your OpenClaw Agent Has No Spending Limit"]]></title>
            <link>https://runcycles.io/blog/openclaw-budget-guard-stop-agents-burning-money</link>
            <guid isPermaLink="false">https://runcycles.io/blog/openclaw-budget-guard-stop-agents-burning-money</guid>
            <pubDate>Fri, 27 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["OpenClaw agents can call any model and retry indefinitely with no budget cap. cycles-openclaw-budget-guard adds hard limits without code changes."]]></description>
            <author>Albert Mavashev</author>
            <category>openclaw</category>
            <category>budgets</category>
            <category>agents</category>
            <category>runtime-authority</category>
            <category>cost-control</category>
            <category>plugin</category>
            <category>tool-limits</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Silent Failures: Why 200 OK Is the Most Dangerous Response in Production"]]></title>
            <link>https://runcycles.io/blog/ai-agent-silent-failures-why-200-ok-is-the-most-dangerous-response</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-silent-failures-why-200-ok-is-the-most-dangerous-response</guid>
            <pubDate>Thu, 26 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Your AI agent returns success, but the answer is wrong. Silent failures are costing teams more than crashes — here's why they happen and the architectural pattern to catch them before damage spreads."]]></description>
            <author>Cycles Team</author>
            <category>agents</category>
            <category>production</category>
            <category>reliability</category>
            <category>observability</category>
            <category>silent-failures</category>
            <category>multi-agent</category>
        </item>
        <item>
            <title><![CDATA["Your AI Agent Just Burned $6 in 30 Seconds — Here's the Three-Line Fix"]]></title>
            <link>https://runcycles.io/blog/runaway-demo-agent-cost-blowup-walkthrough</link>
            <guid isPermaLink="false">https://runcycles.io/blog/runaway-demo-agent-cost-blowup-walkthrough</guid>
            <pubDate>Thu, 26 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A support bot with a quality-loop bug burns ~$6 in 30 seconds. Cycles stops it at exactly $1.00. Three decorators, one exception — zero code change to the agent logic."]]></description>
            <author>Albert Mavashev</author>
            <category>runaway-agents</category>
            <category>demo</category>
            <category>agents</category>
            <category>runtime-authority</category>
            <category>walkthrough</category>
            <category>budget-enforcement</category>
            <category>cost-control</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Runtime Enforcement: Security and Cost"]]></title>
            <link>https://runcycles.io/blog/ai-agent-governance-runtime-enforcement-security-cost-compliance</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-governance-runtime-enforcement-security-cost-compliance</guid>
            <pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Most AI agent deployments lack runtime governance. Pre-execution enforcement — not dashboards — is the missing layer for agent security, cost, and compliance."]]></description>
            <author>Cycles Team</author>
            <category>governance</category>
            <category>security</category>
            <category>compliance</category>
            <category>agents</category>
            <category>production</category>
            <category>MCP</category>
            <category>multi-agent</category>
        </item>
        <item>
            <title><![CDATA["The AI Agent Production Gap: What Developers Are Actually Saying"]]></title>
            <link>https://runcycles.io/blog/ai-agent-production-gap-what-developers-are-saying</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-production-gap-what-developers-are-saying</guid>
            <pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Reddit, Hacker News, and StackOverflow are converging on the same conclusion: AI agents need pre-execution enforcement, not just dashboards. Here's what the community is saying and what it means."]]></description>
            <author>Cycles Team</author>
            <category>agents</category>
            <category>costs</category>
            <category>production</category>
            <category>community</category>
            <category>observability</category>
            <category>multi-agent</category>
            <category>MCP</category>
        </item>
        <item>
            <title><![CDATA["Zero Trust for AI Agents: Why Every Tool Call Needs a Policy Decision"]]></title>
            <link>https://runcycles.io/blog/zero-trust-for-ai-agents-why-every-tool-call-needs-a-policy-decision</link>
            <guid isPermaLink="false">https://runcycles.io/blog/zero-trust-for-ai-agents-why-every-tool-call-needs-a-policy-decision</guid>
            <pubDate>Tue, 24 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Microsoft, Cisco, and OWASP converged on one conclusion: AI agents need zero trust at the tool-call layer. What changed and how to enforce it."]]></description>
            <author>Cycles Team</author>
            <category>security</category>
            <category>zero-trust</category>
            <category>agents</category>
            <category>MCP</category>
            <category>OWASP</category>
            <category>production</category>
            <category>tool-calling</category>
            <category>governance</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Runtime Permissions: Control Actions Before Execution"]]></title>
            <link>https://runcycles.io/blog/ai-agent-runtime-permissions-control-actions-before-execution</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-runtime-permissions-control-actions-before-execution</guid>
            <pubDate>Mon, 23 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Agents don't just cost money — they take actions. Learn why runtime permissions are the missing layer for controlling what AI agents can do in production."]]></description>
            <author>Albert Mavashev</author>
            <category>runtime-authority</category>
            <category>action-authority</category>
            <category>action-control</category>
            <category>agents</category>
            <category>side-effects</category>
            <category>permissions</category>
            <category>engineering</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Action Authority: Blocking a Customer Email Before Execution"]]></title>
            <link>https://runcycles.io/blog/action-authority-demo-support-agent-walkthrough</link>
            <guid isPermaLink="false">https://runcycles.io/blog/action-authority-demo-support-agent-walkthrough</guid>
            <pubDate>Sun, 22 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A support agent can use CRM, notes, and email — but should every run send? Cycles blocks the customer email before execution. Three decorators, one exception."]]></description>
            <author>Albert Mavashev</author>
            <category>action-authority</category>
            <category>demo</category>
            <category>agents</category>
            <category>runtime-authority</category>
            <category>walkthrough</category>
            <category>action-control</category>
            <category>side-effects</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Unit Economics: Cost and Margin Analysis"]]></title>
            <link>https://runcycles.io/blog/ai-agent-unit-economics-cost-per-conversation-per-user-margin</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-unit-economics-cost-per-conversation-per-user-margin</guid>
            <pubDate>Sun, 22 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Model AI agent costs as business metrics — cost per conversation, cost per user, margin analysis — and use budget enforcement to bound variance."]]></description>
            <author>Cycles Team</author>
            <category>unit-economics</category>
            <category>costs</category>
            <category>enterprise</category>
            <category>margins</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["LangGraph Budget Control for Durable Execution, Retries, and Fan-Out"]]></title>
            <link>https://runcycles.io/blog/langgraph-budget-control-durable-execution-retries-fan-out</link>
            <guid isPermaLink="false">https://runcycles.io/blog/langgraph-budget-control-durable-execution-retries-fan-out</guid>
            <pubDate>Sat, 21 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["LangGraph runs pause, resume, retry, and fan out. Enforce per-run and per-node spend limits before durable execution turns cost spikes into cost cliffs."]]></description>
            <author>Cycles Team</author>
            <category>langgraph</category>
            <category>budgets</category>
            <category>engineering</category>
            <category>durable-execution</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["Multi-Agent Budget Control: CrewAI, AutoGen, OpenAI"]]></title>
            <link>https://runcycles.io/blog/multi-agent-budget-control-crewai-autogen-openai-agents-sdk</link>
            <guid isPermaLink="false">https://runcycles.io/blog/multi-agent-budget-control-crewai-autogen-openai-agents-sdk</guid>
            <pubDate>Sat, 21 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Multi-agent delegation chains create recursive cost exposure. Enforce per-agent budget boundaries in CrewAI, AutoGen, and OpenAI Agents SDK."]]></description>
            <author>Cycles Team</author>
            <category>multi-agent</category>
            <category>crewai</category>
            <category>autogen</category>
            <category>openai</category>
            <category>budgets</category>
            <category>engineering</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["Runtime Authority vs Guardrails vs Observability"]]></title>
            <link>https://runcycles.io/blog/runtime-authority-vs-guardrails-vs-observability</link>
            <guid isPermaLink="false">https://runcycles.io/blog/runtime-authority-vs-guardrails-vs-observability</guid>
            <pubDate>Fri, 20 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Runtime authority, guardrails, and observability solve different agent control problems. What each does, where each falls short, why you need all three."]]></description>
            <author>Albert Mavashev</author>
            <category>runtime-authority</category>
            <category>guardrails</category>
            <category>observability</category>
            <category>comparisons</category>
            <category>concepts</category>
        </item>
        <item>
            <title><![CDATA["What Is Runtime Authority for AI Agents?"]]></title>
            <link>https://runcycles.io/blog/what-is-runtime-authority-for-ai-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/what-is-runtime-authority-for-ai-agents</guid>
            <pubDate>Fri, 20 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Runtime authority is the pre-execution layer that decides if an AI agent's next action should proceed — distinct from observability and rate limits."]]></description>
            <author>Albert Mavashev</author>
            <category>runtime-authority</category>
            <category>agents</category>
            <category>concepts</category>
            <category>guide</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Action Control: Hard Limits on Side Effects"]]></title>
            <link>https://runcycles.io/blog/ai-agent-action-control-hard-limits-side-effects</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-action-control-hard-limits-side-effects</guid>
            <pubDate>Thu, 19 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Why controlling what AI agents DO matters more than controlling what they spend — and how to enforce hard limits on emails, deploys, and file writes."]]></description>
            <author>Cycles Team</author>
            <category>action-control</category>
            <category>risk</category>
            <category>agents</category>
            <category>engineering</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["Budget Limits for Claude Code, Cursor, and Windsurf via MCP"]]></title>
            <link>https://runcycles.io/blog/claude-code-cursor-windsurf-budget-limits-mcp</link>
            <guid isPermaLink="false">https://runcycles.io/blog/claude-code-cursor-windsurf-budget-limits-mcp</guid>
            <pubDate>Thu, 19 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Add hard budget limits to Claude Code, Cursor, and Windsurf sessions — config-only enforcement via MCP, zero code changes."]]></description>
            <author>Cycles Team</author>
            <category>mcp</category>
            <category>claude-code</category>
            <category>cursor</category>
            <category>windsurf</category>
            <category>budgets</category>
            <category>agents</category>
        </item>
        <item>
            <title><![CDATA["Manifest vs Cycles: Routing vs Runtime Authority"]]></title>
            <link>https://runcycles.io/blog/manifest-vs-cycles-routing-vs-runtime-authority</link>
            <guid isPermaLink="false">https://runcycles.io/blog/manifest-vs-cycles-routing-vs-runtime-authority</guid>
            <pubDate>Wed, 18 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Manifest optimizes which model handles a request. Cycles decides whether the request is allowed to execute at all. Different layers, different problems."]]></description>
            <author>Cycles Team</author>
            <category>comparisons</category>
            <category>architecture</category>
            <category>agents</category>
        </item>
        <item>
            <title><![CDATA["OpenAI API Budget Limits: Per-User and Per-Tenant"]]></title>
            <link>https://runcycles.io/blog/openai-api-budget-limits-per-user-per-run-per-tenant</link>
            <guid isPermaLink="false">https://runcycles.io/blog/openai-api-budget-limits-per-user-per-run-per-tenant</guid>
            <pubDate>Wed, 18 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["OpenAI org-level spending caps are too coarse for agents. Enforce per-user, per-run, and per-tenant limits before every API call."]]></description>
            <author>Cycles Team</author>
            <category>openai</category>
            <category>budgets</category>
            <category>agents</category>
            <category>per-user</category>
            <category>per-tenant</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Budget Control: Enforce Hard Spend Limits"]]></title>
            <link>https://runcycles.io/blog/ai-agent-budget-control-enforce-hard-spend-limits</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-budget-control-enforce-hard-spend-limits</guid>
            <pubDate>Tue, 17 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Why AI agent cost control must happen before execution — not after — and how the reserve-commit pattern enforces hard spend limits at runtime."]]></description>
            <author>Cycles Team</author>
            <category>budgets</category>
            <category>agents</category>
            <category>engineering</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["Cycles vs LLM Proxies and Observability Tools: Where Budget Enforcement Fits"]]></title>
            <link>https://runcycles.io/blog/cycles-vs-llm-proxies-and-observability-tools</link>
            <guid isPermaLink="false">https://runcycles.io/blog/cycles-vs-llm-proxies-and-observability-tools</guid>
            <pubDate>Tue, 17 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["LLM proxies route calls. Observability tools trace them. Neither enforces budget before execution. See where Cycles fits."]]></description>
            <author>Cycles Team</author>
            <category>architecture</category>
            <category>comparisons</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["Multi-Tenant AI Cost Control: Budgets and Isolation"]]></title>
            <link>https://runcycles.io/blog/multi-tenant-ai-cost-control-per-tenant-budgets-quotas-isolation</link>
            <guid isPermaLink="false">https://runcycles.io/blog/multi-tenant-ai-cost-control-per-tenant-budgets-quotas-isolation</guid>
            <pubDate>Mon, 16 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["One customer's runaway agent can degrade service for every tenant. Enforce per-tenant budgets, hierarchical limits, and scope-level isolation with Cycles."]]></description>
            <author>Cycles Team</author>
            <category>multi-tenant</category>
            <category>budgets</category>
            <category>architecture</category>
            <category>costs</category>
        </item>
        <item>
            <title><![CDATA["Budget Wrapper vs Runtime Authority for AI Agents"]]></title>
            <link>https://runcycles.io/blog/vibe-coding-budget-wrapper-vs-budget-authority</link>
            <guid isPermaLink="false">https://runcycles.io/blog/vibe-coding-budget-wrapper-vs-budget-authority</guid>
            <pubDate>Mon, 16 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A budget wrapper is easy to vibe code. Owning a runtime authority with concurrency safety, idempotency, and multi-tenant isolation is not."]]></description>
            <author>Cycles Team</author>
            <category>architecture</category>
            <category>agents</category>
            <category>budgets</category>
            <category>engineering</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Budget Patterns: A Practical Guide"]]></title>
            <link>https://runcycles.io/blog/agent-budget-patterns-visual-guide</link>
            <guid isPermaLink="false">https://runcycles.io/blog/agent-budget-patterns-visual-guide</guid>
            <pubDate>Sun, 15 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A practical reference for structuring AI agent budgets — covering tenant isolation, workflow caps, run-level limits, graceful degradation, and more."]]></description>
            <author>Cycles Team</author>
            <category>patterns</category>
            <category>budgets</category>
            <category>architecture</category>
            <category>guide</category>
        </item>
        <item>
            <title><![CDATA["AI Agent Cost Management: The Complete Guide"]]></title>
            <link>https://runcycles.io/blog/ai-agent-cost-management-guide</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-cost-management-guide</guid>
            <pubDate>Sun, 15 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["A practical maturity model for AI agent costs — from no controls through monitoring, alerting, soft limits, and hard enforcement with trade-offs per tier."]]></description>
            <author>Cycles Team</author>
            <category>costs</category>
            <category>engineering</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["5 AI Agent Failures Budget Controls Would Prevent"]]></title>
            <link>https://runcycles.io/blog/ai-agent-failures-budget-controls-prevent</link>
            <guid isPermaLink="false">https://runcycles.io/blog/ai-agent-failures-budget-controls-prevent</guid>
            <pubDate>Sat, 14 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["Five AI agent failure scenarios with dollar estimates, and how pre-execution budget enforcement would have caught each one."]]></description>
            <author>Cycles Team</author>
            <category>incidents</category>
            <category>costs</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["The True Cost of Uncontrolled AI Agents"]]></title>
            <link>https://runcycles.io/blog/true-cost-of-uncontrolled-agents</link>
            <guid isPermaLink="false">https://runcycles.io/blog/true-cost-of-uncontrolled-agents</guid>
            <pubDate>Sat, 14 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["What happens when AI agents run without budget limits? Real-world costs, failure modes, and why pre-execution runtime authority is the missing layer."]]></description>
            <author>Cycles Team</author>
            <category>costs</category>
            <category>agents</category>
            <category>incidents</category>
            <category>best-practices</category>
        </item>
        <item>
            <title><![CDATA["How Much Do AI Agents Actually Cost?"]]></title>
            <link>https://runcycles.io/blog/how-much-do-ai-agents-cost</link>
            <guid isPermaLink="false">https://runcycles.io/blog/how-much-do-ai-agents-cost</guid>
            <pubDate>Fri, 13 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["AI agent cost breakdown across OpenAI, Anthropic, Google, and AWS Bedrock — with real-world scenarios for support bots, coding agents, and data pipelines."]]></description>
            <author>Cycles Team</author>
            <category>costs</category>
            <category>agents</category>
            <category>guide</category>
        </item>
        <item>
            <title><![CDATA["Introducing the Cycles Blog"]]></title>
            <link>https://runcycles.io/blog/introducing-cycles-blog</link>
            <guid isPermaLink="false">https://runcycles.io/blog/introducing-cycles-blog</guid>
            <pubDate>Fri, 13 Mar 2026 00:00:00 GMT</pubDate>
            <description><![CDATA["We're launching our blog to share engineering insights, product updates, and best practices for runtime authority in autonomous systems."]]></description>
            <author>Cycles Team</author>
            <category>announcement</category>
        </item>
    </channel>
</rss>