← Gautam Parab

I Read Twelve Agent Codebases in a Weekend. The Dead Ones Taught Me the Most.

Eleven days ago, on 13 August 2026, the maintainers of Flowise archived their repository. Fifty-five thousand GitHub stars, twenty-five thousand forks, one of the defining projects of the 2023 low-code AI wave — frozen, read-only, npm packages deprecated. Their farewell note was unusually honest for this industry: as models got better at reasoning, developers stopped wiring nodes on a canvas and started handing whole tasks to coding agents, and the visual workflow approach simply hit its ceiling.

That announcement is why I spent this weekend reading agent codebases. Not skimming READMEs — actually reading them: twelve repositories, from a 43-star slide-generation demo to AutoGPT’s 186,000-star monorepo. My day job is architecting the governed context that grounds AI agents in enterprise data, so I have a professional obligation to know which of these ideas are real and which are stars-driven theater. What I found surprised me. The frameworks disagree loudly about everything visible — language, licensing, philosophy — and agree almost completely about the things that matter.

The graveyard votes first

Two of the twelve are cautionary tales, and they fail in opposite directions.

AutoGPT’s original 2023 agent — the viral loop that took a goal, self-prompted, and ran until it succeeded or your API budget didn’t — now lives in a folder literally named classic/, kept around for benchmarking. The team’s actual product today is the AutoGPT Platform: agents composed as explicit graphs of typed blocks, with integrations, schedules, and a marketplace. Read that pivot carefully, because it is a confession. The most-starred agent project on GitHub concluded that unbounded autonomy was the bug, not the feature, and rebuilt itself around the opposite premise: reliability comes from structure you can inspect.

Flowise failed from the other side. It had all the structure — every chain and agent a visual node graph — and discovered that structure without capability is just friction. When a coding agent can produce and revise the whole pipeline in conversation, a drag-and-drop canvas stops being an abstraction and starts being a tax. The 2023 thesis was that visual builders would democratize LLM apps. The 2026 result is that the canvas was a transitional interface, useful precisely as long as the models underneath it were weak.

Put those two failures together and you get the actual design space: too little structure kills you, and structure in the wrong layer kills you too. The structure has to live in the contracts — what an agent is allowed to do, what it must produce, when a human signs off — not in boxes and arrows.

Where the structure lives, across three positions Three positions. Structure nowhere: AutoGPT's 2023 self-prompting loop, now kept in a folder named classic. Structure in the interface: Flowise, archived on 13 August 2026. Structure in the contracts: CrewAI, Agno and ZeroClaw, still shipping. A dashed arrow runs from the first position to the third, marking where AutoGPT rebuilt itself. STRUCTURE NOWHERE STRUCTURE IN THE INTERFACE STRUCTURE IN THE CONTRACTS AUTOGPT REBUILT HERE AutoGPT the self-prompting loop, 2023 KEPT IN CLASSIC/ Flowise every chain a visual node ARCHIVED 13 AUG 2026 CrewAI · Agno · ZeroClaw declare · approve · resume SHIPPING
Too little structure kills you; structure in the wrong layer kills you too. Only the third position survived contact with better models.

What the survivors quietly agree on

Here is what struck me most: four production frameworks, written independently in two languages on three continents, have converged on nearly identical answers to the hard questions.

Every task declares its output before it runs. CrewAI — MIT-licensed, 57,000 stars, and pointedly built without LangChain — will not let you define a task without an expected_output field. The validation fails without it. That is a small, deeply opinionated decision: the framework refuses to let you dispatch work you cannot evaluate. Anyone who has managed people recognizes the principle; it is remarkable how long it took agent frameworks to encode it.

Human approval is a paused run, not a pop-up. Agno (formerly phidata, rebranded in January 2025) stores sessions, traces, and pending approvals as rows in your own database, served through its FastAPI-based AgentOS runtime. ZeroClaw — a Rust runtime that ships as a single static binary and claims under five megabytes of memory — defaults to a “supervised” mode where medium-risk operations wait for a human and high-risk operations are refused outright, with cryptographic receipts on every tool call. CrewAI pauses a task on human_input=True. Different stacks, same architecture: the human checkpoint is durable state with an identity, something you can query, resume, or expire — not a modal dialog the agent hopes someone clicks. For anyone putting agents in front of enterprise customers, I would call this the single non-negotiable pattern of the twelve repositories.

Autonomy is a parameter, not a product line. CrewAI’s split between Crews (agents that decide how) and Flows (developers decide exactly what runs when) is the cleanest articulation, but ZeroClaw’s risk profiles and Dify’s workflow engine express the same idea: one execution engine, with the degree of agent discretion dialed per task. Nobody serious ships a separate “autonomous mode” codebase anymore. The graveyard explains why.

Read the licenses, because the business model now lives there. Dify’s license is Apache 2.0 until you read the two added conditions: no multi-tenant operation without written authorization, and no removing the Dify logo from the console. AutoGPT’s monorepo is split — the historic agent is MIT, the platform is Polyform Shield, which permits self-hosting but forbids offering it as a competing service. This is the quiet standardization of 2025–26 open source: the code is open, the SaaS business is fenced. I don’t say that as criticism — it is a rational response to the cloud-provider strip-mining of the last decade — but if you are building on any of these, the license file is now architecture documentation.

The best documentation was never meant to be published

The two strangest repositories on my list are archives of other people’s system prompts. x1xhlol/system-prompts-and-models-of-ai-tools collects the extracted internals of commercial coding agents — Cursor, Devin, v0, Manus — including their tool schemas, and has accumulated roughly 143,000 stars. CL4R1T4S, maintained by the jailbreaker known as Pliny the Liberator, does the same for the frontier chat assistants, framed explicitly as a transparency project. The two were created a day apart in March 2025, which tells you something about when the industry realized the prompts themselves were the product.

Set aside the ethics of extraction for a moment — I have real reservations about republishing commercial prompt text, and I would not paste a line of it into a production system. As engineering evidence, though, these archives are unmatched, because they show what the best-funded teams in the world actually ship rather than what they blog about. Three patterns repeat across nearly every leaked tool: planning, writing, and dispatching are separate modes with separate instructions, never one mega-prompt; asking the user is implemented as an explicit tool, so the model escalates instead of guessing; and the polished artifact the customer sees is kept in a separate channel from the agent’s working notes and citations. None of that is published in any vendor’s documentation. All of it is load-bearing.

Small repositories, sharp edges

The big frameworks supplied the patterns; the small projects supplied the craft.

GenSlide is a 43-star LangGraph demo that turns documents into PowerPoint, and it gets two things right that products a thousand times its size get wrong: it plans the outline before writing any slide, and when a reviewer asks for changes it edits only what the feedback names instead of regenerating the deck. Anyone who has watched an AI tool destroy twenty minutes of accumulated tweaks with a full regeneration understands why surgical revision is the feature.

html-anything, an Apache-licensed local editor out of the OpenDesign team, inverts the usual SaaS architecture: it ships no API keys at all, instead detecting the coding-agent CLIs already authenticated on your machine and spawning them against a library of 75 skills — each skill just a folder with a SKILL.md and hard-coded design constraints (baseline grids, contrast minimums, a ban on placeholder text). Skills as plain files, discovered by convention, versioned in git: I think that idea outlives the editor.

prompts.chat — which most people still know by its old name, awesome-chatgpt-prompts — quietly made the most elegant licensing decision of the twelve: the platform code is MIT, but the prompt corpus itself is CC0, public domain, and remains plain CSV and Markdown in the repository even as the product grew into a self-hostable platform with an MCP server. The data outlives the app. That is how you steward a commons.

And Tencent’s WeKnora, open-sourced in July 2025, is the only project on the list with a serious answer to what agents should produce from a pile of enterprise documents. Not chat answers — a wiki: interlinked Markdown pages the agents maintain autonomously, with the one feature that makes it viable, which is that human edits carry revision history, line-level diffs, and rollback, so an overnight agent run cannot silently trample what a person deliberately wrote. Knowledge as a versioned, co-edited artifact rather than a bag of retrieval chunks is, I suspect, where most enterprise RAG ends up.

The twelve, in one line each

Repository One-line verdict
AutoGPT The pivot from autonomous loop to typed block graphs is the industry’s confession, filed in a folder named classic/.
Flowise Archived 13 Aug 2026; the honest obituary for visual LLM builders.
CrewAI A required expected_output field is worth more than most orchestration features.
Dify Excellent workflow engine; read the license conditions before you build on it.
Agno Approvals and traces as rows in your database — production concerns in the runtime, not bolted on.
ZeroClaw A single Rust binary that treats risk tiers and tool receipts as the core product.
x1xhlol’s archive The coding-agent industry’s real documentation, 143k stars, never meant to exist.
CL4R1T4S The same, for frontier chat assistants, wearing a transparency manifesto.
prompts.chat MIT code, CC0 data: the most graceful commons stewardship of the twelve.
GenSlide Forty-three stars; outline-then-fill and surgical revision done right.
html-anything Zero API keys, 75 skills as plain folders; the skill-file convention will outlive the app.
WeKnora Agents that maintain a wiki humans can safely edit — the best answer yet to “what should RAG produce?”

Contracts, not runtimes

If I compress the weekend into one instruction, it is this: steal contracts, not runtimes. Every one of these projects would love to be your platform, and none of them should be — not because they are bad software, but because the runtime is the part that ages. Flowise’s canvas is already gone; AutoGPT’s loop is a museum piece; today’s frameworks will be refactored beyond recognition within two years. What persists are the agreements: declare the output before dispatching the work; make human approval a durable, resumable state; keep the customer-facing artifact separate from the agent’s working notes; store skills as versioned files; give agent-maintained knowledge the same revision protections you would give a colleague’s.

Those five contracts appeared independently in Python frameworks, Rust binaries, Tencent’s Go services, and the leaked prompts of companies that have never spoken to each other. When that many teams converge under competitive pressure, it is not fashion. It is the shape of the problem showing through — and it is a far better guide to building agent systems than any single framework on this list.

Repository details — stars, licenses, dates — were verified against GitHub as of 24 August 2026.