bitcrowd

Be honest: do you actually know what your legacy system does? Not what the ticket says. Not what the README claims. What it does - edge cases, undocumented behaviour, the fix someone shipped at 11pm three years ago. Most teams don't. And that's exactly why rewrites fail. Next week we're doing a live demo of Surveyor - a tool that turns a real codebase into a verifiable, executable specification. You walk away with an architectural map, behavioral specs per component, and a test you can run on every commit to prove nothing has drifted. Then we open it up for 20 minutes of questions. Bring your messiest situation.

Claude Opus 4.8 introduces dynamic workflows for verifiable outcomes. This shows how AI coding agents have a quiet problem: The bigger the task and the longer the session, the more they drift from what you actually asked for. Strictly speaking, it's not an AI problem. Give a human dev team a huge operation with a loose spec and the outcomes are just as volatile. It's a specification problem. Every product owner dreams of a dev team they can hand their ideas to, and have them magically fill all the unspecified holes in just the right manner. bitcrowd's aim is to be just that team, so we know how hard it is. The key, of course, is to develop the specifications interactively — to arrive at a clear definition of done. A contract for the parts you care about, leaving the team free to decide on the best path to get there. Coding agents have tried to tackle this in various ways. Anthropic's newest approach is Dynamic Workflows: you specify the behavior and the criteria for a correct outcome, and the agents handle the verification. Their showcase: hundreds of agents porting Bun from Zig to Rust, graded against the existing test suite. 99.8% passing, eleven days — the suite was the bar, not a human reading 750,000 lines. It worked because the task was highly verifiable. An executable contract. But how do you turn specifications into contracts? We got there from the other end, building Surveyor — a tool that extracts specifications from codebases. It soon became clear we needed specs that could be turned into a runnable contract. That became Assay: behavioral contracts in plain text you run against a real system, any time someone asks "is this still true?" BDD with teeth. You scope bigger work in more detail — because you're specifying behavior you can check, one Assay specification at a time. We've built the tools. Now we need real-world codebases with real edge cases to make them better. So here's the deal: try Surveyor and Assay on something that matters to you. We'll help you set it up and get working. You pay us with an honest review. Early access is open — comment "spec" or DM us and we'll get you in.

See this already on Thursday at the Elixir Berlin Meetup (16.04.2026)

@elixirberlin is back from the easterbreak on Thursday (16.04.26), 19:00, bitcrowd Talks: Talk about the SSH OTP application by Cara Surveyor - Extracting Specifications from Legacy Software with Elixir by Chris

Read our blog on how to make this work!

RAG is probably the simplest AI enhancement you can have. It stops your LLM seeing things that aren't there, gives you more information and is a good foundation for your AI system. However, it can be hard to evaluate RAG, and many teams just test manually and then push to production. If you want to deploy safely, come to Chris's workshop!

𝐃𝐁 𝐈𝐧𝐟𝐫𝐚𝐆𝐨 𝐧𝐞𝐞𝐝𝐞𝐝 𝐚 𝐫𝐨𝐜𝐤-𝐬𝐨𝐥𝐢𝐝 𝐟𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐟𝐨𝐫 𝐭𝐡𝐞𝐢𝐫 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐁𝐨𝐱 - 𝐃𝐢𝐞 𝐀𝐛𝐡𝐨𝐥𝐬𝐭𝐚𝐭𝐢𝐨𝐧. They chose 𝐄𝐥𝐢𝐱𝐢𝐫 and 𝗯𝗶𝘁𝗰𝗿𝗼𝘄𝗱. Syncing over 1,000 parcel lockers at train stations across Germany, handling offline states, eventual confirmations and ensuring resilience under changing network conditions. Made with Elixir: 𝐂𝐥𝐢𝐜𝐤 𝐚𝐧𝐝 𝐂𝐨𝐥𝐥𝐞𝐜𝐭 𝐡𝐚𝐬 𝐛𝐞𝐜𝐨𝐦𝐞 𝐚 𝐭𝐡𝐢𝐧𝐠. However, customers still need to got to the stores to collect. Home delivery is difficult for commuters, as the delivery windows seldom match their schedule. DB InfraGo had the perfect solution: A service that uses smart lockers to bring click-and-collect to places 𝐭𝐡𝐚𝐭 𝐦𝐢𝐥𝐥𝐢𝐨𝐧𝐬 𝐨𝐟 𝐜𝐨𝐦𝐦𝐮𝐭𝐞𝐫𝐬 𝐩𝐚𝐬𝐬 𝐛𝐲 𝐞𝐯𝐞𝐫𝐲 𝐝𝐚𝐲: The over 5,000 railway stations! The idea for '𝐁𝐨𝐱 – 𝐃𝐢𝐞 𝐀𝐛𝐡𝐨𝐥𝐬𝐭𝐚𝐭𝐢𝐨𝐧' was born: 𝐒𝐞𝐜𝐮𝐫𝐞 𝐥𝐨𝐜𝐤𝐞𝐫𝐬 𝐚𝐭 𝐬𝐭𝐚𝐭𝐢𝐨𝐧𝐬 𝐰𝐡𝐞𝐫𝐞 𝐩𝐞𝐨𝐩𝐥𝐞 𝐜𝐚𝐧 𝐩𝐢𝐜𝐤 𝐮𝐩 (𝐚𝐧𝐝 𝐝𝐫𝐨𝐩 𝐨𝐟𝐟) 𝐝𝐞𝐥𝐢𝐯𝐞𝐫𝐢𝐞𝐬, 𝐫𝐞𝐧𝐭𝐚𝐥 𝐢𝐭𝐞𝐦𝐬 𝐚𝐧𝐝 𝐞𝐯𝐞𝐧 𝐦𝐞𝐝𝐢𝐜𝐢𝐧𝐞 — 𝐫𝐢𝐠𝐡𝐭 𝐨𝐧 𝐭𝐡𝐞 𝐰𝐚𝐲 𝐡𝐨𝐦𝐞! For the InfraGo team, it was clear that this would be a challenging project in terms of the number of stakeholders involved and the technical difficulties of combining real hardware with a digital product. 𝐓𝐡𝐞𝐲 𝐝𝐞𝐜𝐢𝐝𝐞𝐝 𝐭𝐨 𝐭𝐚𝐤𝐞 𝐧𝐨 𝐫𝐢𝐬𝐤𝐬 𝐚𝐧𝐝 𝐜𝐡𝐨𝐬𝐞 𝐛𝐢𝐭𝐜𝐫𝐨𝐰𝐝 𝐚𝐧𝐝 𝐄𝐥𝐢𝐱𝐢𝐫. 𝐀 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: 𝐂𝐨𝐧𝐧𝐞𝐜𝐭𝐢𝐧𝐠 𝐃𝐁 𝐈𝐧𝐟𝐫𝐚𝐆𝐨’𝐬 𝐬𝐞𝐫𝐯𝐢𝐜𝐞𝐬 𝐰𝐢𝐭𝐡 𝐃𝐇𝐋 𝐏𝐚𝐜𝐤𝐬𝐭𝐚𝐭𝐢𝐨𝐧𝐬. The lockers lack a permanent internet connection, yet pickups and deliveries can still happen. When connectivity returns, change events can arrive in bursts and need to be synchronised. Decisions were made: Rather than scaling the team endlessly, the project relied on a 𝐬𝐦𝐚𝐥𝐥, 𝐚𝐠𝐢𝐥𝐞 𝐚𝐧𝐝 𝐡𝐢𝐠𝐡𝐥𝐲 𝐟𝐨𝐜𝐮𝐬𝐞𝐝 𝐠𝐫𝐨𝐮𝐩 𝐨𝐟 𝐞𝐱𝐩𝐞𝐫𝐭𝐬. Working alongside the stakeholders, bitcrowd developed the product, designed the system architecture and built the platform that powers the service. 𝐎𝐧𝐞 𝐤𝐞𝐲 𝐞𝐧𝐚𝐛𝐥𝐞𝐫 𝐟𝐨𝐫 𝐭𝐡𝐢𝐬 𝐰𝐚𝐬 𝐄𝐥𝐢𝐱𝐢𝐫. Elixir has it's origins where small nodes need to cope with high loads, high concurrency and dropped connections. This resillience has made it a go-to language for companies like WhatsApp and Discord. It is also an ideal software stack for Box: When you need to collect your luggage from the locker to catch your flight home, 𝐲𝐨𝐮 𝐫𝐞𝐚𝐥𝐥𝐲 𝐰𝐚𝐧𝐭 𝐚 𝐬𝐲𝐬𝐭𝐞𝐦 𝐛𝐮𝐢𝐥𝐭 𝐰𝐢𝐭𝐡 𝐄𝐥𝐢𝐱𝐢𝐫! Two key takeaways: * 𝐂𝐡𝐨𝐨𝐬𝐞 𝐭𝐡𝐞 𝐫𝐢𝐠𝐡𝐭 𝐭𝐨𝐨𝐥𝐜𝐡𝐚𝐢𝐧 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐣𝐨𝐛. Elixir provided the reliability, scalability and fault tollerance that the project required. * 𝐖𝐡𝐞𝐧 𝐜𝐨𝐦𝐩𝐥𝐞𝐱 𝐨𝐫𝐠𝐚𝐧𝐢𝐬𝐚𝐭𝐢𝐨𝐧𝐬 𝐜𝐨𝐥𝐥𝐚𝐛𝐨𝐫𝐚𝐭𝐞, 𝐬𝐦𝐚𝐥𝐥𝐞𝐫 𝐞𝐱𝐩𝐞𝐫𝐭 𝐭𝐞𝐚𝐦𝐬 𝐜𝐫𝐞𝐚𝐭𝐞 𝐜𝐥𝐚𝐫𝐢𝐭𝐲.

It's happening tonight! Another Elixir Berlin User Group Meetup!!! 18:45: Doors open 19:05: Welcome & Announcements 19:15: 🗣 AI without Neural Nets? An ongoing research project Leo 19:45: break 20:15: 🗣 Privacy by design Lukas 20:45: Lightning talks and Socializing All @bitcrowd HQ in Köpenicker Str. 126 Hurry, you can still make it!

Keep going, Chris!