AI enablement

Getting AI into production

Of course you can show a good demo. The value shows up later: when the system runs on real traffic, holds up on the hard cases, and moves a number the business cares about. That part is engineering and judgment, and it is the part I do, hands-on.

What I actually do

I have built AI products end to end and put them in front of real users and paying customers. With Luminik I run a multi-agent platform in production. With Alfred I run a fleet of coding agents in the open. The same work helps a team go from "we tried some AI" to "AI does this job reliably, and we can prove it."

Concretely, I help with:

  • Find the work worth automating. Start from a task people already pay to do, not from a tool. Pick the few where AI changes the economics.
  • Build the first version fast. A real working slice in weeks, against real data, so the team can feel it instead of debating it.
  • Make it trustworthy. Evals, a calibrated judge, and adversarial tests, so you know how it behaves before customers do. This is most of the work and most of the value.
  • Ship it and keep it honest. Reliability, cost control, and the failure handling that long-running AI needs, so it survives contact with production.
  • Help people use it. Tie it to a number the business already tracks, and bring the team along so it sticks after I leave.

A worked example

I took a vendor support agent into production for a real, customer-facing workload in a regulated industry, and wrote up exactly how we tested it before turning it on and measured it after. It is the clearest picture of how I work.

Read the write-up

Where this comes from

  • Luminik: a multi-agent platform connecting B2B event spend to revenue, $6M+ in sourced pipeline.
  • Aura at Bain: built it as venture CTO, $0 to $3.6M ARR in 15 months.
  • I write about the engineering behind all of this: agents, evals, durable systems, and cost.

Start a conversation

If you are putting AI into production, or hiring someone who has, I am glad to get into the specifics of your problem. The fastest way is a short call; email works too.