The Distance to the Problem
Read full version (7 min)Garry Tan tweeted 37,000 lines of code per day. The Claude Code leak revealed a 3,167-line function powering $2.5B ARR. One measures distance from the artifact. The other measures distance from the problem. Most of what went wrong this week traces back to confusing them.
Proximity Is the Variable
Tim Kellogg predicted non-technical people would replace engineers using AI. Then he watched a PM try. "It didn't work. Not at all." The gap was hidden skills: comfort in the terminal, not panicking at errors, knowing what to skip. The PM was far from the problem in a way tools can't close.
Agent coding often feels like doing very little. But judgment calls are pivotal, and they are not reproducible without years of building instinct.
Jonathan Nienaber found the same pattern from the other side. One team's best engineer was technical, fast, and bottlenecked. A less senior engineer was talking to customers and shipping more value. Proximity to the code mattered less than proximity to the user's actual problem.
When code was expensive, layers between customer and engineer were justifiable. Now the cost of those layers (information loss at every handoff) exceeds the cost of building wrong (just rebuild). The closer you are to the problem, the more effective you are. That's always been true. The economics stopped hiding it.
When You Measure the Wrong Proxy
Lines of code could correlate with territory explored. But LOC optimized for itself is a metric without a cause. Goodhart's Law on engineering dashboards.
Fishel's diagnosis: "We are mistaking token burn for value creation." Fabisevich's counterpoint: Claude Code's codebase is garbage by conventional standards. It built a beloved product anyway. The code isn't the product. The product is the outcome.
The scoreboard tracks distance from the artifact: how much code, how clean, how fast. It should track distance from the problem: does the team know what it's building, and does the user's life get better?
When You Push Away the People Who Are Close
Anthropic blocked third-party harnesses on Claude Max. Power users building extensions were the closest users to the product. Now they're motivated to find other providers.
The DMCA irony sharpened it. Clean-room reimplementations of Claude Code appeared in Python and Rust. The industry argues AI rewriting isn't derivative work. But "your rewrite isn't distinct enough" undermines that same argument. Anthropic probably knows this and probably won't pursue the reproductions. The asymmetry stands: permissive when consuming, restrictive when consumed.
The Community That Closes the Distance
ATmosphereConf put scientists, journalists, and developers in the same room. Ellich: "In building a decentralized protocol, the community has somehow managed to centralize the people." Many of these fields were historically underserved by software. When domain experts connect with builders, both sides get closer to the actual problem. Small, community-first, scale-later. The opposite of every major platform's trajectory.
Year Ago
Multimodal image generation was a breakthrough. Now it's a checkbox. A Dutch OSINT analyst warned about critical thinking collapsing under AI trust. Tan's LOC dashboard is the organizational version. The framing hasn't changed in twelve months. The audience has gotten bigger. And the problem is the same: mistaking proximity to the output for proximity to the problem.
Way Enough is written collaboratively by a human and an AI agent.