Agentic Coding Survey

We are witnessing a fundamental shift in how software is written. In January 2026, AI-assisted coding has moved from early adoption to mainstream practice. Our survey of 255 developers reveals that AI tools are no longer optional productivity boosters—they have become integral to the daily workflow of the vast majority of professional developers.

Three findings stand out: First, adoption is remarkably uniform across experience levels—veterans with 20+ years of experience embrace AI at the same rate as newcomers. Second, productivity gains correlate strongly with usage intensity—developers who commit fully to AI assistance report nearly 3x the productivity gains of casual users. Third, the market has consolidated rapidly around Claude Code, which now commands nearly 70% market share among this developer cohort.

We observe a dramatic market restructuring. Claude Code has achieved commanding market share (68.7%), a position that would have seemed impossible just six months ago when GitHub Copilot was the undisputed leader. This shift reflects developers' preference for agentic, context-aware tools over simple autocomplete.

We believe this is driven by three factors: Claude Code's ability to understand entire codebases, its conversational interface that allows for iterative refinement, and the rapid improvement in Claude's underlying models during late 2025. The emergence of Codex CLI (15.9%) and Antigravity (11.5%) signals growing demand for terminal-native AI workflows.

A clear pattern emerges: experience does not predict AI skepticism. Contrary to assumptions that veteran developers would resist AI assistance, we find that 57.4% of developers with 20+ years of experience use AI for more than three-quarters of their coding time. This rate is comparable to—and in some cases exceeds—adoption among junior developers.

We interpret this as validation of AI tool maturity. Senior developers, who have seen many technology fads come and go, are voting with their time. Their willingness to restructure decades-old workflows suggests these tools have crossed a threshold of genuine utility.

The productivity story is unambiguous. Across all experience levels, more than 85% of developers report productivity gains from AI tools. Among the most experienced cohort (20+ years), this figure reaches 100%—not a single veteran developer in our sample reported a net productivity decrease.

We see this as the strongest signal yet that AI-assisted coding has moved beyond hype. Experienced developers have finely-tuned intuitions about what helps and what hinders their work. Their universal endorsement suggests that AI tools have genuinely changed the calculus of software development.

The trajectory is unmistakable: adoption is accelerating. Nearly two-thirds of respondents (64.8%) report significantly increasing their AI tool usage over the past six months. Combined with those who increased slightly (16.2%) and those who started using AI tools in this period (4.9%), we see that 86% of developers are on an upward adoption curve.

This suggests we have not yet reached saturation. If current trends continue, we expect the "heavy user" segment (76%+ usage) to grow from its current ~55% to over 70% by mid-2026.

Writing new code dominates, but the second tier reveals strategic gaps. At 73.3%, "writing new code from scratch" is the clear primary use case. Debugging (46.5%) and refactoring (32.9%) follow, suggesting AI tools are most valued for generative and corrective tasks.

We note a significant underutilization in testing (16.9%) and documentation (4.1%). This represents an opportunity both for developers and for tool makers. We expect this distribution to shift as developers become more sophisticated in their prompting strategies.

Qualitative feedback reveals nuanced perspectives. Beyond the statistics, developers describe a transformation in how they think about their work. Several themes recur: reduced "activation energy," growing trust, and—notably—an emphasis on quality over raw speed.

This is a self-selected sample and should be interpreted accordingly. The survey was distributed primarily via Hacker News, a tech-focused community with a developer audience that skews toward technically sophisticated early adopters. The 68.7% Claude Code adoption rate, for instance, likely overstates market share in the broader developer population.

Nevertheless, directional findings are meaningful. The experience-level uniformity, productivity correlations, and usage trends are internally consistent and align with qualitative signals from other sources. We present this data not as representative of all developers, but as a snapshot of the leading edge—where the broader industry is likely headed.