Claude

Jason Clinton, deputy CISO at Anthropic, reported in July 2026 that Claude writes about 80% of the code merged at the company, and that an internal instance merges more than half of it. The same model sits at the other end of David Thomas's account of a small graphic animation project: features arrived fast for a week, regression cycles took over in the second, and he reached unmaintainable code in 18 hours spread over five evenings rather than 18 months.

What surrounds the model does the deciding. Katelyn Lesse, who leads engineering for the Claude developer platform at Anthropic, points to a thinking budget configurable in tokens, MCP, a memory tool, context editing, and Agent Skills. Guillaume Dumortier built a marketing system for a 60-person team inside Claude and found that generation was the easy part: six of his seven agents do nothing but check. Clinton's controls run on the same premise, since his incident response agents hold three permissions and cannot deploy.

Positioning against it is constant. GPT-5.6 and Kimi K3 are both priced and benchmarked against Claude, and Ethan Mollick, citing GDPval, records that Claude scores as the better financial adviser while ChatGPT scores as the better sales director. Mollick also finds companies where IT and legal still block tools that regulated competitors already run, Claude Code included.