Pick a category. Pick the business task. We re-weight the underlying benchmarks and redraw competency vs cost. Bubble size = speed. Filled = cloud, outline = open weights.
Same question, two answers. The cloud leader for raw quality vs the open-weights leader for self-hosting.
Better if you want top-tier quality with zero infrastructure, fast setup, and elastic scale — and sending requests to a managed API is acceptable for your data.
Better if data must stay in-house for compliance or privacy, you run steady high volume where self-hosting is cheaper, or you need full control and predictable cost.
A short, no-jargon questionnaire that maps your task, volume, and accuracy needs to shortlist three models with monthly cost estimates.
A short, plain-English read on the week in AI — the decisions worth your time, not the hype.
EP / 002 · 11 Aug 2026Attackers are now using AI to weaponize vulnerabilities in hours instead of days, Kimi K3 are undercutting frontier US models on price, Google's three-tiered AI robotics system, liabilities showing up when AI agents in production go rogue, and more...
"Which model?" is usually the third question. The first two are how much you're willing to build, and how much of your own data you'd put behind it. Here's the spectrum from off-the-shelf to fully bespoke.
ChatGPT for the team, Cursor for the devs, Midjourney for the design lead. Someone else picked the model, wrapped it in a UI, and runs the infrastructure. You pay per seat. You can be productive this afternoon.
You’re embedding a model into a process — support routing, document review, content pipelines. Call a cloud API for raw quality, self-host open weights for control and cost, or do both. Most real systems end up hybrid.
Most teams don’t need this. RAG + a frontier model handles 80% of the cases people think need a custom model. If your domain is genuinely esoteric (legal, medical, niche industrial) and you have labeled data, fine-tuning a 7B–14B open model is the move.
We run AI strategy and implementation engagements for teams that want a sharper edge and less hedge. Engagements start with a 15-minute fit call — short, honest, no pitch deck.
Not every team is a fit. If we're not the right firm for what you're trying to do, we'll tell you in the first 10 minutes — and point you toward someone who is.
The week's verdict. News worth your time, and a chart you'll forward to your team.
