The honest landscape

Most AI advice is written to sell you something. This wasn't.

Get the 20-page field report: which of your workflows can use AI, which can't, and how to make a first move that pays for itself.

Read the report

For the people who actually have to make the AI call, whether that's for a whole company or a single department. Capital is tight, the CFO needs a defensible number, and Europe can't afford another year of deliberation.

Free PDF
Field report · 20 pages

Self-Funding
AI

Where AI cuts cost or does more with fewer people, sovereignly, on hardware you already own.

Malogicamalogica.ai
What you'll walk away knowing

Four things you can act on by tomorrow.

Which of your workflows can legally use a hyperscaler API, and which can't touch one regardless of price.

Which of three proven patterns fits the problem you actually have.

What a first AI move looks like when it's scoped to pay for itself in six months.

Where the real cost of an AI project hides, so the number you take to your CFO survives contact with reality.

What's inside

Eleven sections, written to be useful to someone who has to decide.

  1. 01The first question to ask: can you legally use a hyperscaler API at all?
  2. 02Where API use is genuinely viable, and where it isn't.
  3. 03The break-even math, worked through three real examples.
  4. 04What a self-hosted appliance actually costs, line by line.
  5. 05Why today's API prices are not the prices in three years.
  6. 06The two-speed model lifecycle problem: deprecation and silent drift.
  7. 07A live case: what shifted in three weeks of one frontier model's launch.
  8. 08Where the real cost of an AI program hides.
  9. 09Three architectural patterns that work, and where each one fails.
  10. 10The EU AI Act, practically, for people who have to decide.
  11. 11How to scope a first move that pays for itself in six months.
Who wrote it

Written by someone who has been building enterprise software for 25 years, and running AI infrastructure in production for the last three: deploying open-weights models, writing GPU kernels for non-NVIDIA hardware, and watching European companies either turn AI into measurable value or quietly write off the budget. Some of what's in the report contradicts the published consensus. Where it does, it's because we've seen otherwise sensible projects start in the wrong direction in production, and changed our thinking based on what actually happens after the demo.

The no-pitch promise

We don't pitch. If AI isn't the right answer for your situation, the report says so. If a third-party product fits better than anything we build, we name it.

When you've read it

Read it before your next AI decision.

When you've read it, if it's changed how you're thinking about your AI question, there's a conversation we're happy to have.

The no-pitch promise

A 30-minute diagnostic call. No obligation, no sales pitch. If we can't help, we say so. If a third-party product is the right answer, we name it. We don't pitch.

Book a 30-minute call