datasette code-frequency chart on GitHub

The Trigger: A Chart that Sparked Reflection

Simon Willison, a well-known developer and creator of the Datasette project, is also a blogger who closely tracks AI's influence on programming. Recently, he shared an intriguing image on his blog: the code frequency chart of his Datasette project on GitHub. This chart documents the number of lines added and deleted each week from 2018 to 2026. For most years, the graph remains relatively flat with only minor fluctuations, but in 2026 there is a massive spike — a single week saw 37,022 lines added and 9,528 deleted.

Why is this spike so striking? Because it closely coincides with the release of several top-tier AI models: Anthropic's Opus 4.8, OpenAI's GPT-5.5, and later GPT-5.6 Sol. Simon believes this is no coincidence; it's a tangible reflection of how AI coding assistants have directly boosted his development efficiency.

Deconstructing: How AI Changed an Individual Developer's Output

A closer look at the chart reveals a telling pattern. Datasette is a large open-source project maintained by a single person, where Simon acts as both architect and primary coder. In the era before powerful AI tools, his pace was constrained by personal bandwidth: ideation, coding, testing, bug-fixing — every step demanded time. But with the emergence of advanced AI models, especially those with strong code generation and comprehension, the workflow underwent a qualitative shift.

Now, many repetitive coding tasks can be offloaded to AI assistants, and even some design exploration can be accelerated through dialogue with the models. Simon can invest more time in architectural thinking and solution evaluation, while the concrete implementation is dramatically sped up by AI. The result is a natural surge in output within a given timeframe.

This spike also highlights a key trait of AI tools: they enable developers to iterate quickly. Previously, implementing a complex feature could take days; with AI assistance, a rough result might appear in hours, ready for further iteration. The cost of failure drops, the number of attempts rises, and so does the number of valuable contributions.

Trend Insights: From Incremental to Transformative Change

This phenomenon points to a deeper trend: AI is no longer just a nice-to-have helper but is becoming a productivity multiplier for professional developers. For experienced programmers, AI not only saves time but also opens up new possibilities — like tackling features once deemed too time-consuming or responding to community needs much faster.

Looking more broadly, if more maintainers gain such capabilities, the entire open-source ecosystem could accelerate its evolution. Software iteration cycles would shorten, user needs would be met more swiftly, but new challenges could emerge: increased pressure on code review, technical debt accumulation, or the project becoming overly dependent on a specific AI tool.

Furthermore, the chart hints at a "model-driven" development rhythm: when a new, powerful model drops, developers might enter a period of heightened productivity. This suggests that keeping an eye on AI model progress can itself be a strategy for optimizing personal work patterns.

Practical Takeaways: Partnering with AI

For everyday developers, Simon's experience offers several actionable insights:

1. Leverage AI, but keep control: Let AI handle what it does well (boilerplate code, tests, refactoring suggestions), but always maintain rigorous oversight of code quality and architecture. Don't blindly trust model output.
2. Seize the moment: When you notice a new model has made significant leaps in coding ability, proactively schedule more intensive development tasks to capture the productivity bonus.
3. Focus on creation: The time AI saves should be reinvested in high-value activities — exploring new ideas, learning new domains, improving user experience — rather than mindless scrolling.
4. Prioritize quality over quantity: More lines of code isn't automatically good. It must be paired with stricter testing and review processes to avoid a "speed trap."

Counter‑intuition: AI isn't taking jobs, it's creating more work

Many worry that AI will replace programmers, but Simon's chart offers a different reading: AI didn't replace him; it enabled him to produce significantly more. For an engineer with ideas and design skills, AI behaves more like an on-demand "team," giving an individual the output power that once required multiple collaborators.

Another easily overlooked angle: this explosive growth might be short-lived. After a developer quickly clears their long-held "wish list" of features, output could return to a steady state or shift toward higher quality rather than quantity. What we are seeing may not be the new normal but rather a special expression during a period of adaptation.

In the end, Simon Willison's chart is just a slice of personal experience, yet it acts as a mirror reflecting the new possibilities of human‑AI collaboration in development. It tells us that the tool revolution has evolved from "does it work?" to "how well can we use it?" — and those of us standing on top of these tools should rethink how we work as well.