Gregory Orton

Anthropic's success could be attributed to the fact that they have better product design. They understood the actual business use cases, tweaked accordingly, and built something that works well for the kinds of jobs people are actually trying to do rather than chasing a platonic ideal of "smart" that nobody can agree on how to define.

GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files. And this is why I trust Fable 1000x more.

There's advice making the rounds: replace your CLI args with a single --json payload so agents can use your tool more effectively. Agents think in structured formats. Nested data maps cleanly to JSON. Flat args force awkward conventions. It's a reasonable hypothesis. We tested it. Built a synthetic CLI with 30+ values across nested services. Ran it across five models in both modes. Args swept correctness: 5/5 on every model, including the smallest one in the test. JSON correctness dropped on smaller models. Haiku 4.5 managed 2 out of 5 correct deployments in JSON mode. Cost was the bigger gap. JSON mode cost 4x to 11x more per task across every model. The retries are where the money goes. When the CLI rejects the JSON, the agent reads the error, attempts a fix, and tries again. JSON has more ways to fail, and each failure means more output tokens. The shell escaping tax made it worse. The agent produced correct JSON that broke on shell escaping. Different shells have different quoting rules. On PowerShell, the cost gap between args and JSON was 9x. On Bash it collapsed to 1.5x. Same model, same payload. Args were stable across both: $0.05 vs $0.07. The deeper finding: args constrain the input space enough that even smaller models produce correct output consistently. JSON requires the model to manage its own structure, and that's a capability tax not every model can pay. Keep your args.

AI made me a 100x dev. I’m 100x more frustrated now.