Field notes from the Podesta team on Applied AI — examples-driven workflow building, customer wins, and where the category is going.
The naive view is that since AI lowers the cost of writing code, we should write more of it. Why deliberate over a feature when you can prototype five versions by lunch? The conclusion seems to follow, but I think it gets the economics backwards. Writing code has never really been the bottleneck for most software. Read…
Our developer toolchain runs on open source, self-hosted in our own Kubernetes cluster. Forgejo for git, issues, and PRs; Forgejo Actions for CI; Forgejo's container registry for the images we ship; and CSS of our own on top so the UI reads as a continuation of the rest of the product. All four pieces share a namespace…
We run an internal prediction market. Anyone in the company can open a market on a falsifiable claim about our work, anyone can take either side, the currency is play money, and the leaderboard is visible to everyone inside the company. The point is to invert the usual workplace incentive on disagreement. The default m…
Our CI runs against a self-hosted Forgejo Actions setup in the same Kubernetes cluster as the product. Every job runs inside a microVM (Kata Containers on Kubernetes), not a regular container. Here is why we made that bet, what it cost, and what broke along the way. Why microVM, not container. A container shares the ho…
We run our eval judge three times per case in parallel and report the median. The reason we started doing this was the textbook one — language models are non-deterministic, so replicate the noisy component to defang its noise. The reason we kept doing it is more interesting: in our pipeline, the judge turned out not to…
We build an AI agent for European patent work, and the interface we built for our users to actually talk to it is email. They send a message to a dedicated address, attach the documents their patent process already produced — an examination report from the EPO, a draft set of claims, the prior art the examiner cited — …
Most "X vs Y" posts in this space pick a winner in the first paragraph and back-fill the rest. After shipping production work on all three of n8n, LangGraph, and our own Akribes, the honest take is that there is no universal winner — there are three different shapes of problem, and the tools sort cleanly onto…
We build AI for patent workflows on top of the Podesta platform. Evaluating that agent honestly is the hard part of building it. Patent reasoning is multi-dimensional — the outcome reached, the legal basis cited in support, the route from inputs to conclusion, and how the system handles the cases experts themselves dis…