← All agent-stack verdicts

Agent frameworks

Adopt and wrap Confidence Medium Group Orchestration Evidence as of 2026-09-23

Written by VerdictsOrchestration · Reviewed by control-plane pane 2 on 2026-09-23 (separate AI agent sessions; review record)

Bracketed tags are evidence tiers from RULEBOOK.md §1: [Verified] means checked directly, [CI-observed] read from a project’s CI pages, [Maintainer claim] asserted by the project, [External] a dated third-party source, [Inference] our judgment. A gold path:line link opens the cited line. NODUS is the four-ring verdict scale (Invest, Pilot, Explore, Monitor) used for the FrankenSuite assessments; the agent-stack verdicts use their own scale (Adopt, Adopt and wrap, Build clean-room, Watch). TRL is technology readiness level, scored 1 to 9. CI is continuous integration: automated checks that run on every push.

Bottom line

Inference, medium confidence: adopt an existing agent framework (LangGraph, the OpenAI Agents SDK, smolagents or DSPy) and do not write your own agent loop. Wrap it, because an agent run is nondeterministic and no surveyed framework ships a record-and-replay harness for model calls: script the model and the tools in your tests, make a test fail when the script and the run disagree, and declare any model-graded ("judge") test matrix before you run it.

Adopt, do not rebuild

License checked on 2026-09-23; non-permissive licenses are named in the text.

Copy these practices

Build only if

  • No hard constraint in the evidence rules out every incumbent. The one evidenced gap is a test tool, not a framework: none of the surveyed repositories had a record/replay harness for model calls, and the companion itself marks that absence as contested, so build that harness beside an adopted framework rather than a new framework around it. [Inference] (ecosystem/pickup/_evidence/agent-frameworks.md:40 no project had an obvious record/replay (cassette) harness for LLM calls; ecosystem/pickup/pickup-agent-frameworks.md:141 No surveyed agent-framework repo exposed an LLM-call record/replay (cassette) harness)
  • If you need crash-proof history and deterministic replay of a whole run, that is a workflow orchestrator's job, not a reason to write a framework: the companion assigns durability to orchestrators and bars frameworks from claiming it. [Inference] (ecosystem/pickup/pickup-agent-frameworks.md:24 A framework MUST NOT claim)

Where FrankenSuite touches this

What we cannot say

  • Only LangGraph has verified production adopters; adopter claims for the other frameworks were not checked (ecosystem/pickup/_evidence/agent-frameworks.md:35 Adopter claims for other frameworks are not verified here).
  • microsoft/autogen is not listed above because its last push was 2026-04-15 and its claimed continuation was unverified in the pack (ecosystem/pickup/_evidence/agent-frameworks.md:36 AutoGen is stale-ish: last push 2026-04-15). A fresh read on 2026-09-23 of https://api.github.com/repos/microsoft/agent-framework returned an active repository (pushed 2026-09-23) describing itself as a framework for "building, orchestrating and deploying AI agents and multi-agent workflows" [Verified]; whether it is AutoGen's successor was not established, and its CI and tests were not read.
  • crewAI and agno were checked for test-directory layout only, so no practice is claimed from them and they are not listed as incumbents (ecosystem/pickup/_evidence/agent-frameworks.md:38 Thin test evidence for crewAI and agno).
  • The pack records that its file pointers were fetched and confirmed to exist; it does not record any of these test suites being executed (ecosystem/pickup/_evidence/agent-frameworks.md:39 all file pointers above were individually fetched and confirmed to exist).
  • The record/replay gap is an absence finding over the surveyed test trees only; such a harness may exist elsewhere (ecosystem/pickup/_evidence/agent-frameworks.md:40 That is a gap a clean-room project could fill rather than copy).
  • Stars and same-day pushes show these projects are alive and used, not that they are good (ecosystem/pickup/_evidence/agent-frameworks.md:35 trend evidence is star-count + same-day push activity).

Revisit when

  • An incumbent ships a record/replay harness for model calls, or a surveyed repo turns out to have one: the wrap shrinks to configuration.
  • The AutoGen continuation (AG2 or microsoft/agent-framework) is verified at repository level with its CI read.
  • A third-party agent benchmark publishes its own run-to-run variance, which would decide whether such benchmarks can serve as trajectory oracles.

Sources

Starting a build here?

Begin from the pickup companion for Agent frameworks: it carries the charter seed, the oracle candidates and how to verify them, and the first claims to register, so a new project starts from evidence instead of from scratch.

Help fix it

Each opens a GitHub issue form with this page’s address filled in. You need a GitHub account.

Made by Joshua Nowak (ZestStream) with AI coding agents. Not affiliated with or funded by Jeffrey Emanuel; Joshua uses his public tools and is a paying subscriber to jeffreys-skills.md.

Source on GitHub MIT licensed