How I build

I run my own companies on the systems I recommend. The advice comes from a workflow I use every day, not a slide.

the operation

I do not recommend AI I have not run myself. My own dev work, my own back office, and the products above all run on the same kind of agents I build for clients. When something breaks, I have usually already hit that failure and fixed it.

development workflow

Projects behind my development workflow

Each published panel describes one responsibility and links to the reviewed project example behind it. The panels are independent parts of my working method; no shared runtime relationship is implied.

Agent Config portfolio simulation in a stacked layout, showing a Detailed shared fragment and current Claude Code output followed by the stale Brief Codex output.
image · Portfolio simulation screenshot with Illustrative outputs: the Detailed shared fragment and Claude Code output remain current while Codex shows the Brief output as Out of date. The stacked layout preserves the complete visible excerpts. No real compiler was run and no source stamp was produced.

Shared instructions and tool outputs

Agent Config

Explore Agent Config
Agent Team portfolio simulation at Delivery, showing fictional sample CI and automatic merge, with the default configuration expanded and enforcement settings off.
image · Portfolio simulation screenshot: a fictional successful run reaches Delivery through the represented default legacy automatic merge path. Merge and queue enforcement, sandboxing, and anomaly termination are off; hard caps are unset. The website performs no repository operation.

Review, retry, and blocked work

Agent Team

Explore Agent Team
Control Center portfolio simulation showing a failed run, a stale run, one unresolved decision, sample spend of $18.40, and a fictional deployment record.
image · Portfolio simulation screenshot: the Fails scenario and Needs attention filter show two runs and one unresolved decision, with fictional spend and deployment records. Control Center is a Prototype; deployment is unverified. Ages use a frozen sample time. Homepage and Healthchecks retain credit.

Operator attention and sample records

Control Center

Explore Control Center
Prism portfolio simulation at event six of six, with the result 3 words and an inspect-record excerpt showing two provider turns, one tool call, and completed terminal status.
image · Portfolio simulation screenshot derived from Prism v0.1.0 public source: run.completed, the result 3 words, and a v1 inspect-record excerpt with fixed limits. No Prism execution was captured; measured usage and cleanup receipts are absent from this CLI record.

Event trace and terminal record

Prism

Explore Prism

my stack in practice

None of this is exotic. It is a small set of habits, applied consistently, that add up to a lot more shipped work per week.

primary env
Claude Code as the main development environment
parallelism
agent teams working in parallel on separate tasks
model routing
the right model for each task, not one model for everything
autonomous loops
jobs that keep shipping while I sleep, reviewed in the morning

what this means for clients

Two things. Speed: agents take the first pass on most of a build in parallel while I review and correct, so the work is not gated on one person typing. And judgment: because I run these systems on my own revenue, I know where they break, so I can tell you where AI will not pay before you spend on it.

See the systems

Start with the audit

Audits start at $750, fixed scope. Every engagement begins here. Two to three weeks inside your business, and you end up with a map, a scored shortlist, and one recommended pilot, whether or not you hire me to build it.

Let's talk

Not ready to talk? Take the free AI readiness assessment