Ten seats. Three days. Ship something absurd.
Valon services $140B in mortgages — and quietly runs one of the most aggressive internal AI stacks in any company that isn't a frontier lab. Composer, Claude Code, Codex, frontier models, custom MCP infra, weekly evals. We're cracking it open for ten of you. NYC. July 8–10, 2026.
Boring industry. Absurd AI playground.
Most "AI internships" hand you a Jupyter notebook and a stale dataset. This is not one of those.
Mortgage servicing is supposed to be boring. That's the whole point. Nobody else is here. The legacy stack is held together with Excel and prayer. So we get to build everything ourselves — agents, MCP servers, custom CLIs, internal skills, an open-source platform that lets any agent touch the whole company.
The team is shipping faster than the news cycle. New coding agent drops Tuesday, we've evaluated it by Thursday, we know whether to switch by Friday. Most of you are already living like this. Almost nobody else gets to do it inside a real business with real stakes.
For three days you sit inside that. Same laptops. Same tools. Same problems. Build whatever you can defend on Friday.
Tools up. Everything unlocked.
Whatever harness you like. Whatever model you like. Same internal MCPs, same skills, same evals, same access the team has. You're not in a sandbox.
Coding agents
Claude Code, Cursor Composer, Codex — all wired up. Pick your driver. The team benchmarks them weekly, so you'll know which to reach for by Wednesday.
Frontier models
Opus 4.8, GPT 5.5, Gemini 3.5, OSS. Same model access the engineers have. No watered-down sandbox.
Gestalt MCP
One auth layer. ~20 SaaS tools. Any agent touches all of them. The thing we wrote because nothing else did.
Internal skills
The skills the team wrote because filing tickets was slower than writing prompts. Read them. Fork them. Write your own.
Eval harness
Every new model and harness gets a same-day evaluation. You get the receipts.
Tokens, GPUs, access
Tokens are not the bottleneck. Skill is. Ship until something breaks.
Three days. One demo.
The structure is light on purpose. Enough scaffolding to point you at the right tools and the right people. Otherwise: get out of the way.
Tools up.
Land in the office. Every tool installed and authenticated before lunch. By 4pm you've made your first agent run on the thing you're actually building.
Heads down.
Build. Lunch with the team. Office hours with whoever's shipping that day. One midday check-in. Otherwise: heads down. We protect your time.
Demo & dinner.
Demo what you built to the team. Dinner. Strong demos turn into follow-up conversations with Valon engineering and recruiting.
The self-check.
Be honest. We can tell. Tick the ones that are actually true — three or more and the apply button is for you.
You ship constantly.
Side projects, half-finished products, weird experiments, repos you've abandoned and started again. The receipts exist.
You're already AI-native.
You have opinions about which models and harnesses are good for what. You know what an MCP is. You're not learning the basics on Day 1.
You can explain what you built.
What you made. What broke. What you learned. Without slides, without acronyms.
You bring weird ideas.
You have a running list of things you'd build if you had the time, the tokens, and the right people next to you. We're handing you all three.
You're obsessed with something this month.
Doesn't matter what. Personal knowledge graph, custom agent harness, the new Karpathy video. Specificity over breadth.
Apply. Or send this to the one friend who needs to see it.
Ten seats, closing June 19. Tell us what you'd build. The receipt-having applicants beat the pedigree-having ones every time.
applications & nominations close jun 19, 2026 · 11:59pm ET. paid hourly: $20/hr × 8h × 3 days = $480 total. must be authorized to work in the US without visa sponsorship. under 18 → parental consent required as part of the application. by applying, you consent to photo & video from the program being used in Valon's recruiting and marketing materials. all three days in person at Valon NYC.