> ## Content Index
> Fetch the complete content index at: https://venture-block.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# When Models Ship Your Product: Inside Seattle Startups Where AI Writes Most Production Code
- URL: https://venture-block.com/when-models-ship-your-product-inside-seattle-startups-where-ai-writes-most-production-code/
- Published: 2025-11-01T22:28:19.000Z
- Updated: 2026-10-24T19:42:41.000Z
- Author: Venture Block
- Tags: AI

A recent survey of early-stage, Seattle-area founders sheds light on how deeply AI has penetrated the development process at startups founded in the era of coding-capable models. The poll covered 22 AI-native companies — each with between one and five software engineers — representing 42 engineers in total. The goal was to understand how much of production code (explicitly excluding unit tests, scripts, README files and other non-core artifacts) is being generated by AI tools and agents.

The results were striking. Four of the 22 startups — about 18% — reported that AI writes all of their production code. Another 11 startups (50%) said AI is producing between 80% and 99% of their production code. Taken together, that means roughly 68% of surveyed startups rely on AI for more than 80% of the code that delivers customer value. At the other end of the spectrum, three startups (13.6%) reported that AI writes less than half of their production code.

Survey respondents provided details about which tools they use. Claude Code emerged as the most widely adopted option: 64.3% of the programmers in the sample use it, and 16 of the 22 startups (72.7%) reported using Claude. The average spend per programmer on Claude in September was $167.41\. Cursor, despite its high profile in the press, was used by 23 of the 42 engineers (54.7%), with an average September spend of $113.63 per person. OpenAI Codex arrived in third place in this cohort: seven startups reported using it, accounting for 12 engineers, and average spend on Codex was $48.49 per person in September. GitHub Copilot and Google’s Gemini CLI were less commonly used, representing 9.52% and 4.76% of programmers respectively.

Looking at overall spending on the five most-mentioned AI coding tools, the average per-engineer cost was $182.55 in September, though some startups reported spending more than $400 per person. Founders also listed additional tools in their stacks, including Lovable, Devplan, Mentat, Factory.ai, Jetbrains Junie, Warp and Figma, indicating a diverse ecosystem of assistants and platforms  
contributing to product development.

When asked about barriers to greater AI adoption in coding,  
respondents pointed to several recurring issues. Chief among them was the quality of AI-generated output: engineers often needed to substantially rework what the agents produced. Another common hurdle was the learning curve required to prompt and tune agents effectively. Managing context and working with large, existing codebases was cited as a particularly frequent source of frustration, alongside a broader mismatch between expectations fueled by industry hype and the practical realities of integrating AI into complex projects.

The survey also explored future intentions around tool usage. OpenAI Codex stood out as the tool with the most planned new adoption: nine startups (40.9%) that aren’t using it yet said they intend to adopt Codex in the fourth quarter. When adjusting the data for expected Q4 changes, Claude is projected to retain its leadership position, but Codex would reach parity in terms of the number of startups using it. Cursor and GitHub Copilot show modest downward trends in net intent — each had one startup planning to stop using it — while Gemini CLI could see a small uptick, with three startups planning to try it in Q4.

The findings suggest that AI tools and agents are not competing in a strict zero-sum fashion within these early-stage companies. Today, 68.2% of the startups reported using more than one AI tool to produce production code; respondents expect that share to climb to 86.4% in Q4\. For AI-native founders building companies in an era when models can generate significant portions of production software, the survey offers a window into how tool choice, cost, and the practical limits of current systems are shaping the next generation of engineering teams.