About
I make hidden systems legible.
That is the shortest true sentence I have. The longer version is that I spent fifteen years inside enterprise systems learning why organisations reject software that works, then two years building agentic infrastructure learning what these models can and cannot be trusted to do. The gap between those two is where most enterprise AI is quietly failing, and it is the thing I am most useful for.
The work
Most of my career is ERP: Dynamics 365, integrations, and the unglamorous business of making a finance department and a warehouse agree about what a number means. I managed IT development at National Technical Systems, built manufacturing integrations at Biamp, consulted on D365 implementations at Sikich, and spent three years at Patagonia as a senior developer and technical lead, coordinating delivery across teams in three countries.
The pattern I kept meeting was not technical. The system would work and the organisation would route around it. That is the observation the consulting writing keeps circling: change management is the actual bottleneck, pilots die at integration, and the cost curve nobody models is the one that decides whether a project survives its second year.
I am on a sabbatical now, after moving from Washington to Texas to be closer to family. I have spent it building the thing I wanted to exist: a multi-model system with routing by cost and capability, retrieval-backed memory, durable workflow execution, and an evaluation harness with committed baselines and a regression gate. The evaluation half is the point. Anyone can wire up a model. Knowing when it has quietly got worse is the hard part, and it is the part most teams skip.
How I think about evidence
I would rather have the true version of a story than the satisfying one, and I have learned that those are usually different. The research pieces on this site are built that way on purpose: each claim carries its status, and the writing says plainly which parts the evidence supports, which are contested, and where I am extrapolating past what anyone has shown.
That habit costs me. It makes for worse anecdotes and slower conversations. It is also the only reason anything I say about a system is worth acting on.
What I am useful for
Enterprise AI adoption where somebody needs to say out loud what the models can be trusted with. Evaluation and governance, especially the unfashionable part where you build the harness before you build the feature. Legacy modernisation, where the real work is archaeology. And distributed teams, where the hard problem is rarely the timezone.
If any of that is your problem, the contact form goes straight to me, or email shawntlenker@gmail.com. If you want to know how I think before you write, the second brain answers questions from my own notes, and the writing is all here.