Hey Product Hunt, Santosh here from the CodeAF team. At AgentField AI we've always worked on everything around the model rather than the model itself. Open models are the substrate, our job is the scaffolding that decides how much of their intelligence actually shows up in the output. AgentField started as orchestration for fleets of coding agents. The turn came when we shrank that orchestration down into dedicated harnesses, each built for one kind of job: the same open model produced better work for less money, purely because the harness fit the task. So we benchmarked it. DeepSWE is an open benchmark built from real GitHub issues, run with the official verifiers and the same model behind every harness. CodeAF is currently first out of all popular coding agents by a far margin: · ahead of Claude Code, Codex, OpenCode, Pi, OMP and DeepSeek's own harness · lowest cost per solved issue by far · the one we care most about: beat mini-SWE-agent by a wide margin, and that harness is what most open coding models are RL-trained against, so the benchmark is close to home field for it Then we started using it ourselves at a scale where one chat at a time broke down. We became the bottleneck: watching panes, checking diffs, deciding what runs where. So we built the factory around the harness. You describe the outcome like you'd tell a colleague; CodeAF splits the work, runs the parts in isolated copies of your repo, checks what comes back, and pulls you in only when it needs your judgment. Humans still steer, the software does the middle. One Go binary, Apache 2.0, works with whatever provider you already pay for. What we're pushing next: · Pareto routing learns from every finished run, so each kind of job drifts toward the cheapest model that verifiably handles it (Initial routing model already ships with OS we are improving it and the white paper is in repo as well) · specialists beyond just coding: review, security analysis, then finance and data work outside dev entirely · opening the module that builds specialists, so soon you describe a job and get your own · "Just works across places" we have built and are still building really powerful architecture that makes it just work across devices from devbox to laptops not via ssh, but via actual peer-to-peer transfer with sub second latency, so that you can start your work somewhere, close it and still use it on any other device even if other device is off ! and all Open Sourced. This is a new kind of experience and the UX is not finished. Conversational coding was the right interface when AI was not that smart; as it is exponentially getting smarter, we're still working out what replaces it. Feedback is the thing we want most from this launch, and we've tried to make giving it as easy as possible. If you try CodeAF today, tell us the first moment it feels wrong. Those comments get read first. Though this is a TUI, we will be adding full fledged multi device experience soon.