Claude 3.5 Sonnet — The Agentic Coding Inflection Point
Claude 3.5 Sonnet is released, described as the inflection point model that made long-horizon agentic coding practically viable. First model capable of "sustainable coherent work" across extended sessions — the prerequisite for dark factory development.
Before Claude 3.5 Sonnet, long-horizon agentic coding was theoretically interesting but practically unreliable. Models would lose coherence over extended sessions, make inconsistent decisions, and require constant human correction.
Claude 3.5 Sonnet changed this. It was the first model that StrongDM’s team (and many others) identified as capable of:
- Sustained coherent work across hours-long sessions
- Consistent architectural decision-making
- Reliable multi-file implementation
- Productive iteration on test failures
Why It Matters
The dark factory requires an agent you can trust to run autonomously. That requires a model that maintains context, makes reasonable decisions, and doesn’t randomly drift from the specification.
Claude 3.5 Sonnet hit the threshold where practitioners reported they could assign substantial work and come back to usable results. Earlier models required too much hand-holding to be truly autonomous.
The Downstream Effect
Every subsequent dark factory capability was built on this foundation:
- Claude Code (August 2025) built on the agentic reliability of Sonnet’s lineage
- StrongDM’s factory relies on the agent making consistent decisions autonomously
- The J-curve escape (for teams that achieved it) often dates to when they upgraded to Sonnet-class models
The model release was not the whole story — the tooling, methodology, and organizational changes mattered too. But Sonnet was the necessary (if not sufficient) condition.