The work

Where the method has been tested.

Three places to inspect it: Lucentive's operating base, the regulated-bank engagement the founder leads, and the Intuitive Agent System beta.

Lucentive as operating base

Lucentive runs on the method.

Lucentive is not a slideware firm advising from the outside. The working system is the operating base: repo-local context, agent-assisted drafting, review before publication, and evidence kept close enough that another senior reviewer can inspect how the work moved.

Context travels: Shared product, brand, and workflow context is carried into each run so the team does not restart from a blank prompt.
Review stays in-line: Agent output is treated as work that needs direction, correction, and review, not as finished material.
Evidence remains inspectable: Decisions, diffs, screenshots, and verification output stay near the work they support.

That operating base is deliberately focused. It gives Lucentive a place to prove the method before asking an enterprise team to adopt it.

Field report from regulated-bank production

The method under enterprise load.

A major US regulated bank is where the operating model is tested under production constraints: review, audit, approval boundaries, lifecycle ownership, and engineering work that cannot skip the control path. The founder leads that engagement; Lucentive packages what it learns.

Multi-quarter program the founder leads. AI-assisted engineering shipping through review, audit, and approval boundaries in production. Code generation moved at AI speed; the security review, infrastructure provisioning, deployment approval, and lifecycle steps around it had to be redesigned to match.

Weakest-step diagnosis: The founder walked the seven gating steps of the delivery chain end-to-end and named the one currently capping throughput. The slowest step was not where the program initially focused. Re-sequencing the work against the actual ceiling moved more output than another tooling cycle would have.
Reusable-context layer: The context layer was built for one workflow so it could be loaded unchanged into the next one. From week one, in-line review ran on every agent step and every step produced a record. The setup was packaged so the second workflow ran on top of it without a second design phase.
Lifecycle ownership: Foundation-model updates, context maintenance, policy reviews, and ownership boundaries each got an explicit owner and an explicit cadence. Without that, drift accumulates until the production system stops resembling the design.
Practice capture: Strong-developer practice was captured into shared assets the rest of the engineering organization could run on: when to break out of an agent loop, how to scope a context window, and what an early sign of model drift looks like. Individual leverage became organizational leverage.

The engagement is the founder's; the packaging is what Lucentive does. The four decisions above are written in operating-model terms so the pattern travels into the next environment without naming the first one.

Intuitive Agent System

The proof path in software.

Intuitive Agent System (IAS) carries the context-flow work into software. It gives teams a repo-first harness for reusable context, in-line review (an automated check that runs on every agent step), and evidence on agent-assisted engineering work.

Intuitive Agent System Design Partner program

Live in beta with five Design Partner slots.

Intuitive Agent System (IAS) is the harness side of the work. It runs in your repo, brings reusable context into every run, and keeps in-line review and an evidence trail on every agent step. The Design Partner program is open to enterprise teams shipping AI-assisted engineering in production.

Repo-localReusable contextEvidence trail

Visit ias.dev

Field reports

Read the operating thesis.

Two reads of what is changing in enterprise AI, written from the work.

AI World ThesisTen observations on how AI is changing software production: where leverage compounds, where it stalls, and what the next operating model looks like.

Seven Enterprise ProblemsSeven failure modes enterprise AI programs hit from inside: what is breaking, why the current responses miss, and where to look first.