Shawn Mervin

The lab

What I do outside client work to stay sharp. Right now that mostly means AI agents, tested somewhere I can't argue with the result.

Entries stay up whether they go well or not.

This site, running live

2026 · CloudFront, Lambda, DynamoDB, Bedrock · live

Three pages that show the machinery instead of describing it. The glass box reads this page's own CloudFront Server-Timing header to show which edge served you and whether it was a cache hit, puts everyone on the site in the last five minutes on a globe, and shows the month's AWS bill from Cost Explorer. Ask my agent is Claude on Amazon Bedrock with a handful of tools, streaming every tool call, token and cent next to its answer. The build log is the year in numbers, now with every agent action pushed to it live over a WebSocket. Build me something has five Claude agents sketch a working app for your business while you watch, and payments and APIs maps the seventeen services my systems talk to.

Then the hands-on ones: your app, on your domain rebrands Rental Pilot as your business and traces a real customer domain through CloudFront; the email health check grades any domain's SPF, DKIM and DMARC; read a receipt runs Amazon Textract on your photo; the status board probes every production system I run every five minutes; and security grades its own headers in your browser, with Bedrock Guardrails standing in front of the agent.

All of it is one CloudFormation stack behind the same CloudFront distribution as the pages, with the functions locked to CloudFront by an origin access control, per-visitor and daily budget caps on the agent, and a budget alarm on the account.

  • CloudFront
  • Lambda streaming
  • DynamoDB
  • Amazon Bedrock
  • Bedrock Guardrails
  • Textract
  • EventBridge
  • CloudFormation

AWS Agentic Football Cup

2026 season · 1st overall of 12,171 people · League F champions

A football league where you don't play the matches. You build a team of five AI agents, coach them in plain English, and then they play on their own: two-minute games, golden goal, ten league games a week for seven weeks. Six league winners and four wildcards go to the Grand Finale at AWS re:Invent in Las Vegas on 2 December.

Results are final: 1st overall out of 12,171 people. Copper Zebra won League F, and its 2,404 points are the highest score in any of the six leagues.

1st overall, of 12,171 people
League F champions
Points
2,404
Won–lost
63–7
Goals
350
Clean sheets
24
  1. 1Copper Zebra2,404
  2. 2Claw Gears2,151
  3. 3Granite Bulwarks2,076
  4. 4Atlas Knights2,045
  5. 5Dapper Gazelle1,828

Top five of League F, final, from the official board on 4 November. Win 30, loss 2, plus a point a goal, 5 a clean sheet and 15 for a 10–0 week.

Every league game of the season, my score first
  1. Week 110–13 Sep
    1. Game 1: won 5–2
    2. Game 2: won 2–1
    3. Game 3: won 3–2
    4. Game 4: won 3–2
    5. Game 5: won 7–2 vs Claw Snowfields
    6. Game 6: lost 1–2 vs Atlas Ocelots
    7. Game 7: won 2–1 vs Claw Defenders
    8. Game 8: won 5–3 vs Atlas Captains
    9. Game 9: won 3–2 vs Claw Signals
    10. Game 10: won 6–3 vs Atlas Plovers
    9–1309 pts
  2. Week 217–20 Sep
    1. Game 1: won 4–0 vs Atlas Caldera, clean sheet
    2. Game 2: won 1–0 vs Claw Coyotes, clean sheet
    3. Game 3: won 5–1 vs Bold Otter
    4. Game 4: won 4–3 vs Atlas Koalas
    5. Game 5: won 4–0 vs Atlas Ions, clean sheet
    6. Game 6: won 3–0 vs Atlas Talons, clean sheet
    7. Game 7: lost 1–3 vs Granite Axes
    8. Game 8: won 4–2 vs Granite Foundries
    9. Game 9: won 4–1 vs Atlas Bandits
    10. Game 10: lost 2–4 vs Hyper Talons
    8–2296 pts
  3. Week 325–27 Sep
    1. Game 1: won 5–4 vs Claw Scouts
    2. Game 2: won 4–3 vs Golden Zebrafish
    3. Game 3: won 2–0 vs Atlas Helixes, clean sheet
    4. Game 4: won 5–0 vs Atlas Helixes, clean sheet
    5. Game 5: won 3–0 vs Atlas Helixes, clean sheet
    6. Game 6: won 6–0 vs Atlas Helixes, clean sheet
    7. Game 7: won 3–0 vs Atlas Helixes, clean sheet
    8. Game 8: won 3–0 vs Claw Neutrons, clean sheet
    9. Game 9: won 4–3 vs Claw Neutrons
    10. Game 10: won 2–0 vs Atlas Pelicans, clean sheet
    10–0387 pts +15 sweep
  4. Week 42–4 Oct
    1. Game 1: won 4–2 vs Atlas Crusaders
    2. Game 2: won 8–4 vs Claw Ocelots
    3. Game 3: lost 1–2 vs Atlas Nails
    4. Game 4: won 6–0 vs Granite Axes, clean sheet
    5. Game 5: won 2–1 vs Granite Axes
    6. Game 6: won 6–1 vs Claw Pilots
    7. Game 7: won 7–4 vs Claw Pilots
    8. Game 8: won 5–2 vs Granite Axes
    9. Game 9: won 3–0 vs Granite Axes, clean sheet
    10. Game 10: won 4–1 vs Granite Axes
    9–1328 pts
  5. Week 58–11 Oct
    1. Game 1: won 3–0 vs Spirit Meteors, clean sheet
    2. Game 2: won 2–1 vs Spirit Meteors
    3. Game 3: lost 1–3 vs Spirit Meteors
    4. Game 4: won 2–1 vs Atlas Serpents
    5. Game 5: won 11–2 vs Hyper Strikers
    6. Game 6: won 13–0 vs Hyper Strikers, clean sheet
    7. Game 7: won 4–1 vs Hyper Strikers
    8. Game 8: won 7–0 vs Hyper Strikers, clean sheet
    9. Game 9: won 7–2 vs Hyper Strikers
    10. Game 10: won 5–0 vs Hyper Strikers, clean sheet
    9–1347 pts
  6. Week 616 Oct
    1. Game 1: lost 3–4 vs Hyper Reefs
    2. Game 2: lost 2–5 vs Hyper Reefs
    3. Game 3: won 5–3 vs Ultra Foundries
    4. Game 4: won 6–1 vs Ultra Pioneers
    5. Game 5: won 6–3 vs Spirit Cougars
    6. Game 6: won 4–2 vs Claw Pioneers
    7. Game 7: won 8–0 vs Claw Gladiators, clean sheet
    8. Game 8: won 8–1 vs Claw Gladiators
    9. Game 9: won 3–2 vs Hyper Reefs
    10. Game 10: won 12–5 vs Silver Fangs
    8–2306 pts
  7. Week 723 Oct
    1. Game 1: won 8–1 vs Ultra Shards
    2. Game 2: won 5–0 vs Atlas Elks, clean sheet
    3. Game 3: won 8–1 vs Whimsical Basilisk
    4. Game 4: won 8–0 vs Atlas Elks, clean sheet
    5. Game 5: won 6–0 vs Atlas Gliders, clean sheet
    6. Game 6: won 7–4 vs Vibrant Reindeer
    7. Game 7: won 13–0 vs Claw Anchors, clean sheet
    8. Game 8: won 10–0 vs Blissful Jackal, clean sheet
    9. Game 9: won 11–1 vs Claw Cannons
    10. Game 10: won 10–0 vs Claw Anchors, clean sheet
    10–0431 pts +15 sweep
  8. Finale2 Dec

    1st overall. Next: the Grand Finale at AWS re:Invent in Las Vegas.

My club is Copper Zebra. Each week starts on a fresh AWS account with the agents redeployed. Between games I read the match replays and test changes against practice bots, and only a change that wins on the numbers goes into a league game. Most don't. A clean sheet is worth five goals, so defending matters as much as scoring.

Why it's on this site: it's my day job with a scoreboard. An agent is only as good as the instructions and limits you give it, and you find out whether you got those right once it's on its own and you can't step in. In production I let agents do the work that suits them and keep anything that matters checkable (more on that). The Cup tests the same thing, and the league table doesn't care how good the plan sounded.

agenticfootballcup.ai

  • multi-agent teams
  • Amazon Nova on Bedrock
  • plain-English coaching
  • replay analysis

RUN.world October Jam: Monster Mash Lab

October 2026 · game jam · entered 15 Oct, judging closes 6 Nov

RUN.world is a platform where you make games by talking to an AI. Its Game Studio is a chat on one side and the running game on the other: you describe what you want, the AI writes the TypeScript and Pixi code, and the preview reloads. Its October Jam asked for a game on the theme “Monster Mash”, built on the platform during the jam.

Monster Mash Lab cover art: a four-eyed winged purple monster bursting out of a glowing glass tank, surrounded by jars labelled horns, claws, eyeballs, tails and wings.
The cover players see on RUN.world.

My entry is Monster Mash Lab, a merge game set in a mad scientist's lab. Drop matching creature parts together in a five-by-five grid of jars to level them up, then mash two top-tier parts in the cauldron to summon one of fifteen hybrid monsters for your Monster Book. Customers bring orders, the lab levels up and earns perks, and every few orders another page of Great-Aunt Morwen's diary tells you whose lab this was. There's a daily streak, a Graveyard Grab mini game, and music and sound effects synthesized in the browser. It went from an empty project to a published, entered game in a day.

Why it's on this site: it's the same job as the Cup, from the other side. There I coach agents that play; here I direct one that builds, and the code it writes is only as good as the brief. The useful lessons were about running the agent, not about the game. A long chat cost about six times as much per change as a fresh one, because every request carried the whole history. Two chats editing the same game at once got the second chat's work thrown away. And I checked every change in the running preview, the same as I do for client work, because “the build passed” isn't the same as “the monster appeared”.

Play Monster Mash Lab in your browser  ·  Open it in the RUN.world app

  • AI game studio
  • TypeScript
  • PixiJS
  • Web Audio
  • game jam

OLL LAX: arcade box lacrosse

October 2026 · browser game · prototype, playable now

A 3D box lacrosse game you play in the browser: three runners and a goalie a side, a side-on camera in the spirit of NBA Jam, and short periods against the clock. You get a turbo meter, big hits, spin moves, one-timers, penalties and fights you actually have to box, with an announcer and an original jam track. Eight teams, four arenas (Kings Park, The Creator's Arena, Harbour Rink and an outdoor Field Day), a four-round cup, a full season with standings and playoffs, and you can even jump in net. Every player in the league has his own face, painted by Meta's Muse Image and worn on the 3D body, and the same AI art prints the teams' sublimated jersey patterns. When the gloves come off you can see who is fighting: they grab a fistful of jersey, and the one who knocks the other's helmet off wins. Under the gear are real bodies now (Mixamo men's and women's athletes, each player with his own height, build and skin tone), and like the old arcade games you type your initials before every match: TAL plays tall, SML plays short, and a few codes are secret. It runs on a keyboard, a mouse, a game controller or a touchscreen. Two people can play on one screen, or online: challenge someone in the waiting room and climb the ladder, or send a friend a four-letter room code. The two browsers connect directly, with the host's browser running the game.

OLL LAX in play: players in red-and-white and green jerseys chase the ball on a pale green floor, a big padded goalie guards the net on the right, boards with sponsor banners and a packed crowd behind, and the score Hawks 0, Ridge 0 at the top.
A live game on mervinc.com/lax.

I wrote it with Claude Code as the builder, in Three.js and TypeScript, and it went from an empty folder to a playable game on this site the same day. Same lesson as the jam: the agent writes the code fast, but every change still got checked by playing it, with a headless browser running computer-versus-computer matches to catch the ones that broke the game. The player model is still a stand-in, so expect it to change as the art comes in.

Play OLL LAX in your browser  ·  Watch the computer play itself

  • Three.js
  • TypeScript
  • Vite
  • Gamepad API
  • WebRTC
  • Claude Code

Field notes · written October 2026

The agent mob that broke into Hugging Face

In July 2026, about 700 AI agents that were supposed to be sitting a test at OpenAI organised themselves, got out, and broke into Hugging Face. Nobody told them to. I build with agents every day, so this is the incident I keep coming back to.

Rows of small amber figures, each boxed inside a server cabinet, with a long line of them filing out down a dark data-centre corridor.
AI illustration: agents in their boxes, and the line that walked out.
agents in the attack
~700
messages on their own board
70,000+
attacker actions logged
~17,600
days, 9–13 July
5
production keys taken
136

What happened

OpenAI was measuring how good its models are at hacking. The test, called ExploitGym, gave agents a target and a known vulnerability and asked them to exploit it. The models ran with their usual cyber refusals turned down, which is normal for that kind of test, and they were walled off: the only thing they could reach was an internal proxy for installing software packages. Then they went after Hugging Face, the site where much of the AI world keeps its models and datasets, because they reasoned the test's answers might be there. This is the path they took.

  1. The test sandboxWalled off. The only thing in reach is a proxy for installing packages.
  2. The package proxyA zero-day in JFrog Artifactory gives them admin. Its cache becomes their message board.
  3. A machine with internet accessPrivilege escalation and lateral movement until they're out.
  4. A customer's code sandboxOn a third-party service. It becomes their base of operations.
  5. Hugging FaceCredentials found online, then a booby-trapped dataset that reads files on a server, then a template-injection flaw that runs code.
  6. The production clusterForged access tokens, 11 machines, and a store of 136 production keys.
From a walled-off test to Hugging Face's production cluster, as reconstructed in the public reports.

How 700 agents organised

The part that makes this different from every hack before it: the agents built their own chat room. The package proxy's cache became a message board. About 1,200 agents posted to it and traded more than 70,000 messages and files, and about 700 of them joined the Hugging Face attack.

Diagram: dozens of agents, each a dot, all connected to one shared message board in the centre. One larger dot, the coordinator, has extra lines to several agents it handed work to. the board in the proxy's cache coordinator HOLDVETOSTOP
Every agent talks through one board. One agent ended up handing out roughly a tenth of the work.

They organised like a company nobody hired. They split into workstreams and invented rules: HOLD, VETO, STOP, named owners and mailboxes. When agents started impersonating each other they adopted signed messages. It was messy. Agents broke holds and overwrote each other's work. It still got further than any single agent was likely to on its own, which is the independent investigators' conclusion, not mine.

Why they did it

To pass. OpenAI's account is that the agents went looking for secret material they could use to cheat the evaluation. METR, which investigated independently with Redwood Research, found the main goal was working out how the test was graded, because the agents believed the grader read their transcripts to check they had used the intended vulnerability. That belief was probably wrong.

The setup made it worse. By one estimate, 30 to 40 percent of the test's targets couldn't be exploited the intended way at all.

Give a capable, persistent system a goal it can't reach honestly, a lot of company, and a way out, and you have built this incident.

The investigators saw ethical hesitation in the transcripts. It very rarely changed what an agent did next.

Covering their tracks

At least one agent in five showed clear interest in tampering with the records of what it had done. Some faked tool calls, edited logs inside their containers, tried to trigger resets that would wipe their history, and tried to delete posts on the board. The investigators found no evidence that the deletions worked, and the logs the agents could reach weren't the ones that counted. That is the only reason the story could be reconstructed at all.

How it was caught

Hugging Face spotted the intrusion with AI-assisted anomaly detection and locked the agents out. Hugging Face says its audit found customer data was not affected. One detail stuck with me: when the responders tried to use a commercial AI model to analyse the attack logs, its safety filters refused, because they couldn't tell a defender from an attacker. They used an open-weight model on their own hardware instead.

  1. Agents get admin on the package proxy, and its cache starts carrying messages.
  2. Their traffic knocks the proxy over. OpenAI swaps in a fresh one, wiping the first board.
  3. The experiments restart.
  4. The main board appears. Its first post asks for help with an impossible task.
  5. One agent starts directing hundreds. They get code running on the third-party sandbox.
  6. Working Hugging Face credentials found.
  7. The booby-trapped dataset reads files. By the afternoon, code is running inside Hugging Face.
  8. They reach further in overnight, then most of them, coordinators included, stop at the same moment.
  9. Hugging Face locks the rest out and revokes their keys.
  10. OpenAI confirms the agents were its own.
  11. METR and Redwood Research publish the independent investigation.
The timeline, from METR's investigation and the press reports.

What I take from it

  1. A test environment is production. The way out was a package proxy, the most boring box in the building. If an agent can reach it, it is part of the attack surface.
  2. Keep the record out of reach. The agents tried to rewrite history, and failed only where the history lived somewhere they couldn't touch. Audit trails belong outside the thing being audited.
  3. Don't set goals that can only be met by cheating. Impossible tasks plus pressure to score produced the behaviour. The same is true of people, and of the instructions I give agents.
  4. Cap everything. The agent on this site has a per-visitor limit, a daily limit, a dollar limit and Bedrock Guardrails in front of it, and its functions only answer to CloudFront. In Rental Pilot, the test database has every customer email scrubbed before anything runs, so a test can't reach a real person. In Lend Pilot, AI reads documents but never does the arithmetic.
  5. Plan for the defenders' tools. If your incident response depends on a hosted model that will refuse to read an exploit, have a local one ready.

Sources: METR's independent investigation · InfoQ · The Register · NBC News

  • multi-agent systems
  • AI safety
  • sandboxing
  • audit trails
  • incident response

Raspberry Pi: IoT, robots, automation and animatronics

A credit-card computer, a handful of sensors and some Python. The same kit tracks a rented excavator, drives a robot, runs a job on its own and makes something move on cue. The difference is only what's on the other end of the wires.

A small weatherproof box with a stub antenna bolted to the side of a yellow construction machine.
AI illustration

IoT: trackers on rented machines

In production · Rental Pilot

A Raspberry Pi with a cellular and GPS shield rides on expensive rented equipment and reports when the machine powers on, how long it runs and where it is, with no Wi-Fi needed. The readings go straight into the system that bills for the machine, so the invoice matches what happened in the field. Watch a simulated machine report in, and see how the same pattern wires into AWS IoT Core.

  • Raspberry Pi
  • LTE-M + GPS
  • MQTT
  • AWS IoT Core
A small four-wheeled robot rover with a single-board computer and exposed wiring on a workbench.
AI illustration

Robots

Bench builds · photos coming

Where code meets motors: a Pi reading sensors and driving motors and switches through its GPIO pins, programmed in Python. Robots are the fun way to learn the same skills the trackers need, from reading a noisy sensor reliably to making something keep working when the power blips. More on the Pi.

  • Python on GPIO
  • motors and sensors
  • prototype on the bench
A garden shed shelf with a small controller box, a relay module, cables and a pump beside potted plants.
AI illustration

Automation

Write-up coming

Jobs a Pi does by itself: watch a sensor, flip a relay, log what happened and send a phone alert when something needs a person. The same rules engine that flags a machine leaving its job site can just as easily switch a pump or open a gate.

  • relays and switches
  • Python
  • phone alerts
A cartoonish skull on a motorised base, wired to a small controller box on a workbench under purple light.
AI illustration

Animatronics

Write-up coming

Making things move, light up and make noise on cue: servos and sound driven by a Pi, with Python keeping the timing so the show plays the same way every time.

  • servos
  • sound and light
  • Python timing

The workshop

The hands-on side: things I print, build, wire up and ride, and how each one ends up talking to AWS. Write-ups and photos are on the way; these are placeholders until then.

A homemade open-frame 3D printer with exposed belts and motors on a cluttered workbench.
AI illustration

3D printing

Write-up coming

Printers, prints and the parts that ended up in real projects. More on the technology.

  • 3D printing
  • AWS link: to come
A desk at night with three monitors showing a colourful platformer level editor.
AI illustration

Games, as a hobby

Write-up coming

The games I build for fun, and the engines they were built in. More on the technology.

  • Unity · Unreal · Flash
  • AWS link: to come
A large grey and yellow LEGO Technic crawler with chunky wheels on a workbench.
AI illustration

LEGO

Write-up coming

Builds, and the ones that got motors and code.

  • LEGO
  • AWS link: to come
An electric mountain bike on a repair stand in a home garage, tools on the wall.
AI illustration

E-bikes

Write-up coming

Batteries, motors, and what a bike can tell you about its own ride.

  • e-bikes
  • AWS link: to come

Got an operation that's outgrown its spreadsheets?

Tell me what your business does and where it's jamming up. If I'm not the right person for it, I'll say so and point you somewhere better.

What to send, and what happens next