"Coding Is Solved" Is a Lie – Anthropic's Flickering Bug
- https://www.youtube.com/watch?v=zfYsSFY4l18
- Original title: I Think They Are Lying To You
Passionate rant against the "coding is solved" narrative pushed by Anthropic and Boris (creator of Claude Code). Prime argues the claim is demonstrably false and actively harmful to developers' mental health, using Claude Code's own year-long terminal-flickering bug as proof that even pure software problems remain hard.
The claim under attack
Boris (Anthropic, Claude Code creator) says he no longer prompts Claude — he writes loops that prompt Claude until a win condition is met, burning huge token volumes. His broader claims: "coding is solved," "coding is the easy part," the hard parts are infrastructure, hardware, and user feedback. Anthropic also tweeted Q2 2026 ships 8x code per employee vs pre-2025 — i.e. "two years of code per employee per quarter."
Why Prime calls it lying
By "lying" he means either intentional deception or such strong self-belief they reject contrary evidence. His exhibit: terminal flickering.
- Claude Code released to researchers Feb 2025; GitHub issue #392 (screen flickering) filed within two weeks, April 11 2025. Many duplicate issues followed.
- Dec 17 2025: Anthropic announces rewritten rendering reducing flicker "by roughly 85%" — Prime mocks an 85%-effective bug fix and the fact they essentially built a game engine to render terminal text.
- Dec 18 2025: rollback of changes "to stay stable heading into the holidays."
- April 1 2026 (>1 year later): Boris ships feature-flagged "no flicker mode" (uses alternate screen buffer, like Vim).
The core argument
Flickering is a pure software problem — no hardware or capacity excuse. If coding were truly solved, you wouldn't need feature branches, feature flags, or year-long bug timelines. Other ongoing failures: "fewer mysterious error messages" promoted as a May 2026 feature, elevated error rates on status page, repeated reports of users seeing other people's session output (possible session leakage they can't reliably fix).
The harm
Prime says the narrative burns out developers: people yeet unreviewed AI code into production, feel their jobs are meaningless, and get anxious about the future. Salespeople with free unlimited tokens preach $10k/day workflows while making false claims. Acceleration of code output is real and fine to celebrate — the dishonesty is claiming coding is solved.