InfluxDB creator Paul Dix argues AI agents will end hand-written code Other

Paul Dix, founder and CTO of InfluxData and creator of InfluxDB, argues in an essay that writing code by hand and reviewing it line by line is "headed for extinction". The essay, "The end of programming", is free to read on his blog at pauldix.com.

Dix opens with a chart GitHub published in its August 20, 2026 post on the August 17 outage. It shows merged pull requests per month at about 130M, commits per month at about 2.9B and new repos per month at about 24M as of August 2026.

The essay was published on August 25, 2026. Dix's claim is that most software will soon be produced by AI agents, and that humans will review the end result, not the code itself. He expects "more software than we've ever had", and also "a mountain of slop".

What it is

• The main example. Dix points to Bun 1.4, released on August 20, 2026, which rewrites the JavaScript runtime from Zig to Rust. Bun creator Jarred Sumner did the port with a pre-release version of Claude Fable 5. Bun's own write-up says Bun was acquired by Anthropic in December 2025 and that Sumner works at Anthropic.
• The numbers, per Bun. The rewrite pull request added 1,009,257 lines and removed 4,024. It took 11 days (May 3 to merge on May 14, 2026) and 6,778 commits. Sumner ran about 50 dynamic workflows in Claude Code, with 64 Claude instances at once. Bun says the tokens used before the merge would have cost "around $165,000 at API pricing".
• The "it had a reference" objection. Critics say a port is easy because the old code acts as an oracle. Dix's answer: "If you can build a verification system and give proper direction, AI can produce a highly complex, highly sophisticated piece of software."
• His own tests. In a fork of InfluxDB, Dix says Fable built an Apache Iceberg integration (API, CLI, compactor changes, Iceberg manifests, an S3 API) in 14 hours. He wrote the architecture and requirements, then had Fable work through sub-agents and code reviews. A second feature, edge-to-central-cluster data replication, reached a mostly working version in 28 hours. Fable then deployed it to test infrastructure in AWS and fixed bugs from logs and metrics. Dix says it is "still running a few months later".
• The limits he admits. Neither feature has shipped and Dix does not call them production ready. He says he cannot afford to run a frontier agent on a 24/7 improvement loop, which he estimates at "hundreds of thousands of dollars" of monthly spend.
• AI labs as a preview. Citing what Anthropic and OpenAI developers post on X, Dix says they "ship dozens or hundreds of PRs per week" and no longer review every line. He expects the rest of the industry to get that experience "a year from now".
• His timeline. By the end of 2027, he expects Fable-level intelligence to be cheap enough for most developers to use full time. In "another year or two", he expects broad access to systems that produce tokens at "10x to 100x the speed at a fraction of the cost", pointing to Cerebras and OpenAI hardware results.
• Not overnight. Dix expects organizational inertia to keep hand-written, human-reviewed code going for "another decade" at many companies.
• Who wins. In his view, the most productive people will direct AIs and build harnesses, "software factories" and QA and verification systems. The key questions become "what do you want to ship? And what do you want to support?"

Bun rewrite in numbers

Sources

• Paul Dix: The end of programming
• Bun blog: Rewriting Bun in Rust
• Bun blog: Bun 1.4
• GitHub: Rewrite Bun in Rust (pull request #30412)
• GitHub Blog: The August 17 outage, and the work ahead