01
Leadership
Change begins with leadership, not with IT. We help boards and executive teams make AI decisions that show up in the results — and stay in the 30 % that succeed.
Every process has steps that must stay predictable. And steps where AI brings more than rules can.
We help you draw the line before it forms by accident — so the transformation delivers outcomes, not one of the 70 % that fail.
Starting point
70%
of AI transformations fail.
The reason is rarely the technology itself. It’s the conflation of a deterministic process with non-deterministic AI — and the wrong tool ending up in the wrong place.
- 74 % trust human judgement over automation
- 68 % keep processes under ten steps
- — reliability is the main challenge, not the model
Source: UC Berkeley, survey of 306 AI professionals
How we work
Leadership, technology, people — all three at once.
AI transformation is not a technology project, an IT project or an HR project. It is all three, running in parallel. When one falls behind, the whole thing collapses with it.
02
Technology
A deterministic process layer and a non-deterministic AI layer, connected under one architectural view. Not a set of point tools — a managed whole that scales into production.
03
People
Skills and culture at the centre. Technology alone is not enough — you need the people who can use it the way the architecture intends.
Where architecture fits
The line runs through the middle of your stack.
Without an architectural view, the boundary between predictable process and AI judgement gets drawn by accident — and the wrong things end up on the wrong side.
Outcome Business outcomes
01 · Deterministic Process layer Workflows · rules · audit trail · same input → same output
— The line Latentti draws —
02 · Non-deterministic AI layer Agents · LLMs · judgement · probability · human-in-the-loop
Foundation Data & integrations
Both layers exist. The question is always per step: which side does this belong on?
How an engagement runs
Six steps, no surprises.
- 01 Identify processes and decision levels
- 02 Set business goals
- 03 Map data sources and integrations
- 04 Redesign processes
- 05 Implement automation and AI agents
- 06 Monitor and iterate continuously
Open work
For AI tools to produce reliable output, they need deterministic context.
Two MIT-licensed templates we use in client work. Free to fork, adapt and contribute back.
Latentti/business-context-ai-template MIT
TOGAF-based business context for AI development tools.
Three tiers — Quick Start (4–6h), Full (8–16h), Enterprise (16–40h). Includes a Claude Code agent for guided completion. Describe your business once, reuse across every AI project.
View on GitHub Latentti/ea-ai-context-template MIT
Enterprise architecture rules for AI tools.
Architecture principles, patterns and ADRs as YAML. Tells AI agents how your organisation builds software — so the output matches your standards, not the model’s defaults.
View on GitHubPull requests welcome. These templates evolve with our client work.
Talk to us
You meet the people who do the work.
No pyramid of analysts, no pre-sales layer. For implementation we partner with your teams or our trusted network.
