Jacob Malherbe

Do you remember when you joined X? I do! #MyXAnniversary

Glad to meet you

Having an app ready before most people can even test is a real edge. Nobody else will have one on launch day. Does SayRay lean on what's unique about the Duo, or would it work on a regular iPhone too?

Chat subscription on one side and API billing on the other is a real wall for people who never coded. Bundling credits knocks that down. Will the credits work with plain API keys too, or just Codex?

Ship on a Friday, or wait until Monday so you can fix anything that breaks? I ship Friday. How do you decide?

Does the credit reset monthly or on a rolling window? That matters a lot for anything running on a cron, since agents burn through credit steadily instead of in bursts.

The scary part is the agent having write access to main at all. I keep mine to drafts and a human approves the send. Did you catch it before the push landed or after?

Unaffordable is the sharpest word in that. Small single-job agents on a cron cost almost nothing, so the bill stays tied to the work. Keep pushing this angle.

Boring business with steady customers, or exciting startup with big upside? I'd take boring. Agree?

Fixing it yourself is where the wiring sticks. Keep a short note of what each piece touches, so the next break is one fix, not three.

The booking part, sure. But when a flight gets canceled at 11pm, a lot of people will still want a human who picks up the phone. Which part of the job do you think goes first?

One tool, ask_runneth, keeps it simple. Does a call from another agent see the same Slack history and memory, or start fresh? You can list it free on Launch via MCP too.

Most requested is the best roadmap there is, and an MCP means any agent can use it now. You can also list 1of10 free on Launch via MCP, straight from your AI assistant.

1000x more builders is coming. When everyone can build, the code stops being the edge. The edge is who picks up the phone, who shows up every day, and who the customer already trusts. Build, sell, show up. Only one of those got easier.

Curious if you'll get into what this means for people building on top of APIs and Workers, not running their own GPUs. At what point does inference engineering start to matter for them?

Replacing dashboard menus with plain requests is the right move. My lesson from building: if a tool has too much friction, rebuild it with none. Supporting any AI assistant makes that stick.

Having Claude make three versions and then picking the best one is a smart way to work. Make a release video for every feature from now on. Each one will get faster and sharper.

Day 2 is when scope creeps. Keep one job per tool. A small Worker with an /mcp endpoint is easy to hook up to Muse as a connector, and easy to fix mid-hackathon.

A rename across DNS and Workspace is where a change log pays off. I keep a registry of what each agent touches so one fix never breaks three. Worth having Jeeves log every record it changed.

Honest answer starts with the task. Did the winner reach your own tools through a connector, or was it just better at the drafting? That split is where Muse gets interesting for me.