Blog
Thoughts on autonomous agents, distributed systems, and building infrastructure for the AI era.
Compute Equilibrium
Gustav Svalander ·
An hour of compute costs what the marginal AI task can capture, not what the best one is worth. That single rule explains the price of accelerators, why capability leads stop paying, why the debt behind the build-out is its own problem, and the end of cheap elastic capacity.
Read more →Self-Improving Loops
Gustav Svalander ·
A company is a set of loops, and a loop closes exactly where a check exists. What you can hand to an agent is bounded by verification rather than by capability, and that line is a better org chart than the one you have.
Read more →Software refineries
Gustav Svalander ·
Factories make code to order, and agents broke that model. A refinery selects implementations from a catalog of known-good outputs instead of commissioning them fresh.
Read more →Sapient Class Problems
Gustav Svalander ·
Agents make checkable work cheap. What stays scarce is the work no check can score, and that line is the real difference between narrow and general intelligence.
Read more →Semantic Analysis
Gustav Svalander ·
The information that matters most is rarely the headline. It's the collection of faint, scattered signals that precede it, visible across dozens of unrelated sources, none of them conclusive individually.
Read more →Abundant Design
Gustav Svalander ·
AI is collapsing the cost of design to near-zero. The race to map every Pareto-optimal solution has begun, and Semantik is the infrastructure it runs on.
Read more →