AI EngineerVideoSource Linked

Agent Frameworks Considered Harmful — Rémi Louf, .txt

Reliable background agents need runtime primitives more than elaborate framework abstractions: durable events, queues, versioned inputs, causal traces, diffs, and replay.

AI Engineer · Aug 22, 2026

Open Source Open Markdown Open JSON

Source Summary

A two-week background-agent experiment exposed duplicate Slack posts, a lost voice note, and an untraceable prompt regression. Those failures drove an **append-only event log**, proper attempt handling, causal links, and a **content-addressed request graph**.

Practical Implication

Keep agent definitions reviewable and separate from runtime machinery. Persist every event and exact model input, version prompts and tools, and make runs diffable and replayable so model swaps and regressions can be evaluated against prior requests.

Agent-Ready Context

A two-week background-agent experiment exposed duplicate Slack posts, a lost voice note, and an untraceable prompt regression. Those failures drove an **append-only event log**, proper attempt handling, causal links, and a **content-addressed request graph**.

Keep agent definitions reviewable and separate from runtime machinery. Persist every event and exact model input, version prompts and tools, and make runs diffable and replayable so model swaps and regressions can be evaluated against prior requests.

This is one builder’s workload rather than a framework comparison. The runtime required custom engineering, and the claim that local or open models were adequate applies to these briefing and note-processing tasks, not necessarily coding.

Connected Context · Feed7 Judgment

This makes replayability and causal provenance concrete requirements for background-agent runtimes: append every event, preserve exact model inputs, distinguish attempts, and version prompts and tools. It narrows the framework critique to failures observed in one workload, while providing infrastructure needed to trace regressions and compare model changes without mixing reviewable agent definitions with execution machinery.

From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, ArizeThe trace-driven improvement loop depends on the exact persisted requests, causal links, and versioned runtime state introduced here; otherwise investigated regressions may not be reproducible.fighting slop with slop — Vaibhav Gupta, BoundaryBoundary relies on transcripts and execution traces for assurance, while this Signal specifies the append-only, versioned data model needed to make those records diffable and replayable.Don't Build Agents You Can't Answer For — Addy OsmaniIt supplies concrete runtime evidence for answerability: exact inputs, attempts, and causal links can support the logs and explicit accountability required before shipping agent changes.Twin: Playing an Unknown Game with a Test-Time Digital TwinBoth treat replay as a reliability mechanism, but Twin validates actions against an executable simulator whereas this Signal uses replay primarily to trace failures and compare runtime or model changes.

Context Map

agentcoding#harness-engineering#observability#agent-reliability

Uncertainty

This is one builder’s workload rather than a framework comparison. The runtime required custom engineering, and the claim that local or open models were adequate applies to these briefing and note-processing tasks, not necessarily coding.