Agents
A five day app build
Five days.A real app.
Agents make it fast. I make it delightful, robust, and better than AI left to itself.
The important distinction
You are not hiring an agent.
You are hiring the person directing all of them. AI supplies extraordinary implementation leverage. I supply the architecture, product judgment, taste, edge case obsession, and final accountability.
Max
Decides what belongs.
Chooses the product shape, rejects plausible nonsense, and makes every piece cohere.Agents
Cover the happy path.
Produce working flows quickly enough to leave real time for scrutiny.Max
Attacks the awkward states.
Loading, failure, recovery, accessibility, copy, performance, and the details AI calls done too early.The standard behind the speed
Fast is the new part. Judgment is not.
I designed Homebrew in 2009. It grew into software used by about 50 million people, while current public analytics record more than 278 million formula installations in a year.
That scale belongs to the community that has carried Homebrew forward. The founding lesson is the one I bring to your app: make the system simple enough to understand, extend, test, and hand over.
The point is not that a five day app begins at Homebrew scale. It is that I know the difference between a demo that works once and software people can trust.
50M
people, approximately
TEDx Savannah source
278M+
formula install events in the latest public year
Homebrew analytics
A record of finishing the job
Proof is in the awkward states.
Delight is not decoration. It is what happens when information, behavior, recovery, and platform conventions have all been considered together.
Automic Vault
Security needs more than a happy path.
A native Mac app that finds hazards, explains them, and gives people a clear route back to safety. The dangerous states are part of the product, not an afterthought.
Visit Automic Vault
Package Manager Manager
Five days in the open.
The first five days of work produced 27 focused commits from me and Codex. They covered testable scanning, progressive loading, main thread discipline, stable rows, native Tahoe chrome, resize stability, and new package manager support.
27commits in the first five days
Canopy
One product across every layer.
Mac and iPhone apps, a server, GitHub webhooks, and push notifications. Delight survives only when the whole system agrees.
See CanopyWhat the five days contain
A compressed senior product team.
The scope is deliberately sharp. That gives us enough time to make the complete core experience real, then make it feel considered.
-
Day 1
Find the product.
Reduce the idea to its essential job, settle the interaction model, and cut everything that steals focus. -
Days 2 and 3
Make the whole loop real.
Direct agents across the interface, data, APIs, platform integration, and tests until the app genuinely works. -
Day 4
Try to break it.
Attack edge cases, failure paths, accessibility, performance, and the seams between systems. -
Day 5
Make it feel inevitable.
Refine typography, motion, copy, feedback, empty states, and the hundred small decisions that create trust. -
Handoff
It is yours.
You leave with the working app, its source, a clear account of what was built, and a sane route forward.
The useful constraint
Focused enough to finish. Serious enough to matter.
Five days works because we protect the core idea from the usual drag. I will tell you plainly if the app needs a different shape or more time.
A strong fit
- One clear job for the app
- One person empowered to decide
- Access to the required APIs and accounts
- A narrow first release worth making superb
Needs a longer runway
- An entire marketplace or social network
- Unresolved stakeholders or product strategy
- A regulated backend beginning from zero
- A platform migration disguised as one feature
Bring me the app in your head