The public notebook.
Education, research, marketing, and engineering in one feed, tagged so you can follow a thread. Formal papers live on the Research page; these are the working notes.
AI agents need boundaries, not babysitters
The useful question is not 'can we trust the agent?' but 'what have we told it it may do?' Bounded agents beat clever ones.
Evidence before execution: why research precedes roadmaps
The cheapest time to learn you were wrong is before you build. A short research pass saves quarters of engineering time.
Marketing that compounds beats marketing that spikes
Campaigns end; systems stay. The marketing worth funding is the kind that still works the month you stop paying attention.
What adult learners actually buy (it is not content)
Adults do not pay for information. They pay for structure, feedback, and a credible finish line. Course design should start there.
Type safety is a business decision, not a style preference
Typed code costs a little more to write and pays back every single week after. The invoice for untyped code arrives as incidents.
Data minimization, in practice, for teams that store everything
'Collect less data' is easy to agree with and hard to do. Here is the four-question routine that makes it operational.
The second audience for your website is an AI
More purchase journeys now start in a chat window. Sites written only for humans are quietly losing the first impression.
Designing courses for people with jobs, kids, and 40 minutes
Completion rates in adult education are dismal everywhere. The fix is not more motivation, it is respecting the shape of a working week.
Why a services company publishes its research
Giving away thinking looks like a cost. For us it is the whole business model: published research is how quality gets proven in advance.
Small teams, big systems: engineering discipline at startup scale
Startups fail from process bloat as often as from chaos. The discipline that matters is a short list, enforced ruthlessly.