Labs Blitz Pack
Product
Blitz Pack, an NFL card game
Platforms
iOS and Android
Stack
React Native, TypeScript, Supabase
Time
100 days, Jul 5 – Oct 13, 2026
Shipped
627 merged pull requests
Team
Six engineers, with product, design and delivery
Try it
Drag the seal across the pack, or tap Rip it. A Hall of Fame pack holds four player cards and one power card, and every Tuesday each player rips three packs: twenty player cards and four power cards for the week. The players here come from one real week-four matchup.
The idea · July 6
Blitz Pack started as a two-page brief from inside the shop: a trading card game first, fed by fantasy data. Every card is a real NFL player, and its power each week is that player's real score.
The brief went after the parts of fantasy that feel like work. No draft, just a sealed pack. No waiver wire: pick a position, rip, and live with what comes out. Power cards played live during the games, so Sunday is something you play rather than something you check.
It was a good brief. Nobody knew yet whether it was a good game, and the only way to find out was to build it and play it.
Version one
A small team built it with Claude and a set of agents from the first day: the backend, one app for iOS and Android, and the design system the agents built from.
Week 1
The first commit already had Claude as a co-author, and it came with a written spec and a plan. The agents built from specs like it all summer, one per feature, 54 of them in July alone. The app was running on a test server by July 16.
Week 3
All 42 components in Figma were linked to their code, so an agent building a screen used the exact colours, type and spacing instead of guessing. When the team moved to a new component library on July 22, every screen was rebuilt on it in a single day.
Week 7
The beta opened on schedule on August 18. Two days later the real card art arrived: every card a real NFL player, scored each week on what that player actually did.
Week 10
Version one reached production on September 8. The next day the whole shop was playing it against each other, with scores updating during the real games.
A month of play
By mid-September the hard part was done: a working game on both platforms, real NFL data, and a production environment running itself every week. And the people playing it every day kept arriving at the same habit. Set the lineup on Wednesday, check the score on Monday.
A fixed date was close, and the usual answer is to ship what works and improve it later. The team decided a card game that plays like fantasy football was not the product, and that the build was fast enough to change the game instead of polishing it.
The turn
Six decisions came out of those two days. Each one fixed a problem the first version made visible, and each one cost something.
A season-long deck made early luck permanent, and a deck already set gave you nothing to do midweek.
A combined score is fantasy football's scoreboard, and it made the cards interchangeable.
Cards committed before Thursday meant the week was decided before a ball was kicked.
With lineups visible, the best move was to copy or counter what you could see.
Decimal PPR scores read like a spreadsheet, not a card game.
Rarity was doing two jobs at once, rating a player and making a card scarce.
Nine days
The people decided what to build and wrote it down. Claude and the agents built it. Everyone played each build as it landed.
Day 0
Four people took two days to rewrite the game: weekly cards, five duels instead of a points total, power cards played live during the games. The plan gave the rebuild until October 16 to be working on stage.
Day 2
The front end was split into fourteen tickets, each one saying what done looks like and which checks it had to pass, so an agent could finish it without stopping to ask. The first one deleted the old game within the hour: trading, the binder, the card index, fifteen screens.
Day 3
The app was rebuilt in two days against stand-in data, so nobody waited on anybody. The real rules landed underneath it by September 30: whole-number scoring, five duels, live power cards, a hidden week. Lint, type and test checks ran on every push, and across the project an automated reviewer read 207 of the 627 pull requests.
Day 9
On October 1 production was rebuilt for the new game, iOS and Android together, with the animated pack rip landing the same day. The deadline was still two weeks away, and the shop spent them playing it.
Then and now
Drag the line. Version one's live duel as designed in August, scored to one decimal; version two's in October: five duels, real card art, whole numbers.
Version oneVersion twoPower cards
A real week-four matchup from production: down one duel to three, with one tied. In the game you play power cards while the games are on, and each one is aimed at a duel. Pick a card, pick a duel, and try to turn it around.
Down 1–3, 1 tied. Two plays left.
Tyler Shough17QB25Joe Burrow
Javonte Williams26RB0Jordan Mason
Nico Collins23WR23Nico Collins
Noah Fant0TE5Juwan Johnson
Brandon Aubrey10K14Jake Bates
TE · Art withheld: our audit flagged this card for the wrong team.
Pick a card from your hand.
100 days
Each bar is a week of commits. Pick an era to see what changed in the game and in the way the work got done.
The turn
Peak: 180
Commits per week, Jul 5 – Oct 13, 2026
01 · Jul 5 – 10
The game
The brief became a first rule set: a sealed starter pack, an eight-card field, one power card a week, and a lock on Thursday.
How we built it
The first commit was a database schema, a project guide written for agents, and a design spec, co-authored with Claude. Licensing questions and the first power-card ideas were researched with Claude too, then argued over by people.
02 · Jul 5 – 21
The game
Clubs, invites, sealed packs, a weekly settle and a live NFL stats feed: the rules ran in the database before there was much to look at.
How we built it
Every feature started as a dated spec and an implementation plan. July alone produced 54 of them. The first screens were plain and functional, and the first Android run opened on a network error.
03 · Jul 17 – Aug 18
The game
The app took its first shape: four tabs, card art, rarity tiers and the pack rip.
How we built it
The front end moved to Tamagui, and every screen was rebuilt on it in one day. Figma components were linked to code through Code Connect, and a build skill told agents how a screen is made in this repository, so a ticket and a Figma frame were enough to build from.
Internal beta started on August 18, the date set in July.
04 · Aug 18 – Sep 21
The game
A season-long deck, an eight-card field, PPR scoring to two decimals, around forty-five power cards locked in before Thursday, rarity on the card face, a binder and trading with a club veto.
From late August the team played a full NFL week every weekday.
How we built it
Production went live on September 8 with the weekly loop on a schedule: lock warnings, autofill, live scoring during NFL windows, settle and next week's packs, with no one pressing a button.
05 · Sep 17 – 23
The game
Playing it every day made the problem plain. The cards read as player avatars, rarity changed no decisions, and by Wednesday there was nothing left to do. It was becoming a fantasy app with a card skin.
How we built it
The schedule said ship. The team chose the game instead: a small group took two days to rewrite the rules, and wrote down what to cut, what to keep and a three-week plan to get there.
06 · Sep 22 – Oct 1
The game
Fresh cards every week, five head-to-head duels, whole-number scoring, power cards played live through Monday night, and an opponent's week hidden until kickoff. Trading, the binder and the archive were deleted.
How we built it
The front end didn't wait for the backend. It rebuilt the game against stubbed data in two days, removing 15,788 lines on the way, and the backend landed the new rules over the next four. On October 1 production was rebuilt for the new game, two weeks ahead of the plan.
07 · Oct 2 – 13
The game
A new design library folded the deck into the Duel screen. After the first weekend of real games came a rulebook, ghost teams for empty seats, and power cards that aim at a duel rather than a player.
How we built it
The redesign tickets were filed and merged the same day. Weekend playtest notes became thirty tickets by Monday, and the week of October 5 took 180 commits, the most of any week.
By the numbers
Commits
1,234
Merged pull requests
627
Commits co-authored by Claude
418
Of 879 non-merge commits. A floor: squashed merges drop the credit.
Unit tests
2,400+
Plus 38 end-to-end harnesses against a real database.
Decision to production
9 days
From the rules rewrite to version two live.
Platforms
2
iOS and Android, shipped together.
Contact