Notes from the infrastructure we ship.

Architecture decisions, compliance posture, Decision Traces, and lessons from production inside a regulated enterprise VPC.

• 
  Jun 22, 2026·9 min read

  The second signer: why regulated AI needs two humans on one decision

  Every AI vendor says their system puts humans in the loop. A second signer is the mechanism that makes that claim auditable, queryable, and defensible in a regulated workflow.

• 
  Jun 20, 2026·9 min read

  Why the demo worked and the pilot didn't

  The demo impressed because a solutions engineer curated perfect context by hand. The pilot failed because a retrieval pipeline replaced that assembly. The model was the same.

• 
  Jun 19, 2026·7 min read

  AI recruiting software made screening faster. It did not make it predictive.

  AI recruiting software automates resume and skill screening. Measured against four years of production data, those signals did not predict who performed on the job.

• 
  Jun 17, 2026·8 min read

  Shadow evaluation: how a model earns its way into production

  Shadow evaluation runs a new model against live inputs with no downstream effect. The step most AI vendors skip is the only one that counts as evidence.

• 
  Jun 16, 2026·9 min read

  What an AI council should ask every vendor

  AI councils in regulated enterprises run vendor reviews on procurement checklists. These six questions expose architecture instead, with what a real answer looks like.

• 
  Jun 15, 2026·7 min read

  Orchestration is what makes thirteen agents one system

  Multi-agent AI fails when the agents don't coordinate. The orchestration layer keeps context, routes work, and makes the system proactive. Without it, you have tools.

• 
  Jun 14, 2026·9 min read

  The gap in the System of Intelligence thesis

  a16z named the right category. The thesis leaves one constraint unstated. For regulated enterprise, that constraint determines whether a deal closes at all.

• 
  Jun 13, 2026·9 min read

  Why six AI hiring vendors got rejected at the same carrier

  A Fortune 500 insurance carrier rejected six AI hiring vendors in eighteen months, every one on architecture. Why that pattern repeats at every regulated enterprise.

• 
  Jun 12, 2026·8 min read

  Agents that can't act alone can't cascade

  DeepMind just funded research into agent populations interacting unpredictably. The enterprise answer is architectural: agents that cannot act alone cannot cascade.

• 
  Jun 12, 2026·6 min read

  Your HRIS is the friend graph. Keep it.

  Facebook's friend graph never died. People just stopped going there. Your HRIS is headed the same way, and that is better news for the CHRO than it sounds.

• 
  Jun 10, 2026·8 min read

  'We need AI' is not a problem statement

  Specific business problems ship; ambitions stall. How to turn 'we need AI' into a scoped, costed problem statement, and why most enterprise AI projects fail.

• 
  Jun 10, 2026·8 min read

  Automate the grunt work first

  Leaders automate the easy 20% and declare victory while the administrative load stays. What to automate first with AI, and why hard processes need a partner.

• 
  Jun 10, 2026·8 min read

  Model quality stopped being the bottleneck. The context layer is.

  Model quality stopped being the bottleneck. What enters the context window, in what structure, decides reliability. The context layer is the durable moat.

• 
  Jun 10, 2026·8 min read

  Why a 34-day deployment reads as a red flag

  Unexplained speed implies skipped controls. How templated connectors, a canonical glossary, and confidence bands make a 34-day deployment inspectable.

• 
  Jun 10, 2026·8 min read

  Evidence before the next talent decision

  Governed outcome evidence shows likely outcomes, risk, upside, uncertainty, and supporting evidence before a named human makes a talent decision.

• 
  Jun 10, 2026·8 min read

  What control model lets you move that fast?

  What control model lets you move that fast? Queryable Decision Traces, logged approvals, and customer-defined second-signature rules answer the speed fear.

• 
  Jun 10, 2026·8 min read

  Approved weight artifacts can leave. Your production data does not.

  How approved weight improvements can move across a governed boundary without moving customer records, outcomes, or decision history.

• 
  Jun 10, 2026·8 min read

  The internal candidate your systems can't see

  Internal mobility fails on rigid taxonomies that hide adjacency. A context graph computes who could make the adjacent move, and humans approve every one.