Positions we're
willing to defend.
Not a blog. Arguments about where software production is going, what breaks on the way, and what it costs to find out late. We publish the reasoning so it can be disagreed with.
Right about the bottleneck. The gate has to scale.
Anthropic is right that code is no longer the bottleneck. The next step is explicit, policy-driven delegation at every lifecycle transition.
Read the articleThe constraint was never in the prompt
Anthropic is shrinking prompts. Solario's path shows the deeper shift: from prose into typed, validated and executable software contracts.
Read the articleWhat makes an agent loop converge
Analysis of 7,565 Qwen3.6 turns identifies action collapse and reference-surface eviction as a measurable production failure mode.
Read the articleWhy remediation doesn't scale
At machine volume, correcting software stops being surgical and starts being regeneration, and regeneration doesn't converge.
Read the articleGeneration is becoming free. Production is not.
Every month the cost of producing a line of code falls. The cost of knowing whether you should have produced it does not.
Why governance compounds
Constraints, decisions and approved exceptions are the only assets in an AI estate that appreciate.
The build reflex
Almost every large engineering organisation is currently building its own agent platform. Most are measuring the wrong thing.
Software factories replace software projects
The project was an accounting artefact of labour-based delivery. When capacity is provisioned rather than recruited, the unit of planning changes.
SPRINT 0 · TWO WEEKS · FIXED SCOPE
Find out which loop you're actually in.
A factual inventory of an estate or a pilot: architectural violations, pattern deviations, undocumented decisions, and the provenance gaps an auditor will find first. No methodology, no commitment, just the measurement.