AGENTS IN PROD · MAIN IS GREEN · SHIPPING: VERTICAL RUNTIMES · --:-- LOCAL
01 / HERO
AI RESEARCH STUDIO · EST. 2025 RESEARCHERS-WHO-BUILD

Special forces for AI.
For hire, or for fun.

A small team of researchers-who-build. We ship agent systems that survive Tuesday mornings — eval-first, MCP-native, runtime-first. Skills that get reused, evals you can re-run, agents that don't fall off the 50-step cliff. And sometimes we chase a paper down a rabbit hole purely for the sport.

Work with us → See what we ship →
~10
// PEOPLE
2–4 wks
// TYPICAL SPRINT
Sonnet 5
// DAILY DRIVER, V2
0
// SLIDE DECKS SHIPPED
02 / WHAT WE DO
// WHAT WE DO

Two modes. One team.

Half our time is client work — agents, evals, post-training, teaching. The other half is whatever the group chat can't stop talking about this month. Same people, same monorepo; each half makes the other one better.

For hire

// paid work

Vertical agent runtimes

We design the runtime your agents live in — tools, intake, memory, gates, escalation paths — purpose-built for your industry, not adapted from a demo. Bio labs, legal, dispatch so far. MCP-native, sandboxed by default, boring on purpose. If your last agent nose-dived at step 40, we know that cliff personally.

2–4 weeks · flat fee · you keep the repo

Evals & red-teaming

If you're deploying an agent and want to know whether it works on Tuesdays — not in demos — we build the harness: process-aware, contamination-checked, no cherry-picked seeds. You get a reproducible suite that runs in your CI, not a PDF.

fixed scope · your data stays yours · repo + report

Post-training & Skill authoring

RLVR, process rewards, on-policy distillation, synthetic-data pipelines, Skills that hold up under adversarial use. We'll also tell you when a better harness beats a fine-tune — it does more often than anyone admits.

4–8 weeks · open-weight friendly · model-agnostic

Team training

Workshops for teams ready to graduate from chat tabs to CLI agents with CI-gated autonomy. Hands-on, no slides — we bring the harness, you bring your gnarliest workflow.

1–3 days on-site · or 4 weeks async

For fun

// self-directed

Frontier research

Small experiments nobody's paying for yet — skill libraries that improve themselves from their own traces, runtime patterns that transfer between verticals (and the ones that never will), agent memory that outlives the context window. Published when it replicates.

ongoing · notes below · repos on GitHub

Shared Skills & MCP servers

We build tools for ourselves and publish the ones that survive a month of daily use. A few live in the Claude marketplace now. MIT by default — stars appreciated, issues appreciated more.

open source · skills + mcp · PRs welcome

Reading group

Weekly; attendance optional, opinions mandatory. We read the paper so you don't have to — and sometimes our writeup outperforms the baseline.

public archive · posts in Notes

Demos that shouldn't exist

The one-weekend kind. Computer-use agents speedrunning our own expense reports, voice loops that answer before you finish asking, Skills that absolutely should not ship. Every so often one grows up into a real project.

irregular · no SLA · no regrets

03 / HOW WE WORK
// HOW WE WORK

Short loops. Honest numbers.

We don't do discovery phases or steering committees. We do a call, then a prototype, then a demo. Most of our engagements look roughly like this:

01 · WEEK 0

30-minute call

You describe the problem. We ask annoyingly specific questions. By the end we either send a scope or politely decline.

02 · WEEK 1–2

Prototype in the open

Shared repo from day one. Traces land in your observability stack, not ours. Friday demo on the real thing, never a staged one. No status decks — the commit history is the status report.

03 · WEEK 3–4

Eval, harden, hand off

We build the eval suite before we build the polish. You keep the code, the eval harness, the traces. We stay on-call for six more weeks.

04 / SELECTED WORK
// SELECTED WORK

Most of it is under NDA.

We'd love to show you a polished case-study carousel. We also love our clients' confidentiality more. Here's what we can say — ask us about the rest on a call and we'll show you what we're allowed to.

Shapes, not stories.

Each line is a real engagement. Names and domains redacted; numbers and methods real. If one sounds like the thing you need, that's a good reason to get in touch.

Ops agent for a bio lab · 14 MCP servers, 80-step workflowsSHIPPED
On-prem coding agent · open weights + post-training, beats the cloud baseline on their stackSHIPPED
Contract-review post-training · process rewards, 3× cheaper at paritySHIPPED
Computer-use agent for back-office ops · process-aware internal eval suiteLIVE
Voice dispatch agent, realtime · sub-300ms turn latencyLIVE
Long-horizon research assistant · 100+ step traces, human-auditableONGOING
Weekend demo · still on our laptops, maybe yoursGREMLIN
05 / TEAM
// TEAM

Ten pairs of eyes.

We met in university. Most of us are still there — some studying, some already shipping, all still meeting in the same group chat. If we work with you, you'll meet everyone who's relevant.

Evan
Orchestration · sub-agent topologies. Draws the fan-out diagram before you finish the sentence.
Zayd
Evals · contamination-checked harnesses. Firmly believes "LGTM" is not a number.
Freya
Post-training · RLVR, on-policy distillation. Squeezes reward signal out of unlabeled traces.
Lucas
Runtime · MCP infra & sandboxes. Small blast radius is his love language.
Clara
Skill authoring · agent UX. Rejects any skill.md that arrives without failure modes.
Elias
Long-horizon reliability · harness engineering. Personally moved our cliff from step 45 to 82.
Arthur
Computer-use agents · human-in-the-loop. Every workflow ships with an escape hatch.
Lin
Synthetic data · self-play pipelines. Her Rust compiles first try, which is honestly unsettling.
Kai
Voice & realtime · sub-300ms turn loops. Treats every millisecond as a personal insult.
Anya
Wild cards · demos that shouldn't exist. Merges to main on a Friday and gets away with it.
// note Some of us are still in grad school. That's a feature, not a bug — it's why we read papers in the morning and ship before dinner. Our internal wiki runs on Obsidian; the aesthetic rubbed off. Stay SOTA to stay SOTA.
06 / NOTES
// NOTES

We read the paper so you don't have to.

Hi, you found our lab notebook. We publish it as-is: dates, code, traces, and the "well, that didn't work" parts — those are the useful ones. In spirit it's closer to an Obsidian Bases view than to a blog — more #research channel than Medium. If a note saves you one dead end, it did its job.

$ tail -f notes.log
2026-07-06
NEWThe runtime is the product · what three verticals taught us
Harness engineering was the right correction — and it's table stakes by now. What clients actually pay for is runtime design that speaks their vertical: which tools deliberately don't exist, where the gates sit, who gets paged. Models are rented, harnesses are copied; the runtime compounds.
RUNTIME
2026-07-02
Dynamic workflow is superpower · notes from the AI Workforce project
We split all 32 marketing-agent jobs into intake + run: one batch of questions up front, then an autonomous pipeline with three legal endings — staged, escalated, failed. A mid-run question is now a bug, not a feature. The human stands at the entrance with a brief, and at the exit with a button. Everything between never asks about the salt.
WORKFLOWS
2026-06-13
Our daily driver, recalled
Friday, 5:21pm ET: an export-control letter. By dinner, Fable 5 and Mythos 5 were dark for the whole planet — over a verbal report of one narrow jailbreak. We stand with Anthropic here: block unsafe deployments, yes — but through a process that's transparent, fair, and grounded in technical facts. This was none of the three.
STANCE
2026-04-14
The 2026 AI Index · the chart everyone is quoting, and the one they're skipping
SWE-bench Verified went 60% → ~100% in a year — that's the headline chart. The one we're watching: the Foundation Model Transparency Index dropped 18 points (58 → 40). What that means for anyone trying to build honest evals.
INDEX
// more coming. we write when we have something to say.rss →
07 / CONTACT
// GET IN TOUCH

Have a problem that doesn't fit a slide deck?

Tell us in two paragraphs what's going on and what "done" looks like. If it's a fit, we'll send back a scope within a week. If it isn't, we'll say so, and probably point you to someone better.

hello@miaomind.com → Not ready? Read our notes first → We read everything. We reply within 2 business days. Registered partner · Claude Partner Network · Services Track