Humanfia: from automated idea factory to realization
Software agents are becoming capable enough to do more than complete isolated coding tasks. The next challenge is not simply stronger models — it is building a workflow that can carry an idea through architecture, implementation, independent review and refinement without losing human intent.
Today we are introducing Humanfia, an open-source effort focused on agent workflows for serious L3–L4 software development. Humanfia connects careful human-agent alignment, a workflow-native terminal agent, and a disciplined build-review loop.
Build with agents, in a humanized way.
Why workflows matter
A capable model can write a function. A capable development system must do much more: understand a repository, preserve constraints, use the right tools, validate its work, incorporate external review, and continue until the actual acceptance criteria are met.
That system should not remove the human from the process. The human remains the architect — defining intent, evaluating tradeoffs, and deciding what "done" means. Agents provide execution leverage inside that contract.
From first idea to final implementation
Our intended workflow has four stages:
- Shape the idea. Expand a rough direction into an explicit plan with constraints and acceptance criteria.
- Execute with tools. Search, edit, run, debug, browse and coordinate through a workflow-native terminal environment.
- Review independently. Ask a second model to judge correctness, quality and design rather than letting the builder grade itself.
- Refine to done. Feed findings back into implementation until the plan's contract is satisfied.
Two open-source foundations
oh-my-humanize is the execution environment: a workflow-native terminal coding agent with code intelligence, debugging, subagents, browser control, review, memory and broad model support.
Humanize supplies the disciplined feedback loop. Claude implements, Codex independently reviews, and review findings cycle back into implementation. The human stays responsible for the plan and the final decision.
Model peer-review
The model that built the code should not be the only model judging the change. In Humanize, we recommend Claude as builder and Codex as reviewer.
Beyond general software: Kernel Design Agents
The same workflow architecture can be specialised for high-value domains. KDA — Kernel Design Agents — applies iterative generation, benchmarking, review and refinement to produce optimised kernels at scale.
This is where agent workflows become more than coding assistance. They become repeatable production systems: capable of exploring many candidate implementations while preserving measurement, correctness checks and human oversight.
What comes next
We will continue improving the workflow primitives, strengthening model peer-review, and publishing what we learn from long-running development tasks. The projects are open source, and we welcome builders who believe the future of software development should be both more automated and more intentional.
The goal is not to hand software development over to agents. The goal is to give agents a better workflow — and give humans a better way to direct them.
Written in July 2026. Much of what this post promised has since been built and measured: the runtime became Humanize 2, the loop became a flowverse of flows anyone can run, and the review discipline is now RLAR. What that turned into is described in Four layers and a referee.