4/ We originally built this to help us move faster while building Demux https://t.co/UEZJC8cfLT
As we use it on real features, we’ll share more about the architecture, what works, what fails, and where coding agents still need humans in the loop.
1/ We wanted to spend more time on the actual engineering side of a task:
understanding the problem, designing the solution, defining constraints and tests.
Then delegate as much of the implementation as possible to AI agents.
So we built a small coding harness for it.
3/ Each job gets a fixed budget.
The harness clones the repo into an isolated environment, runs the different stages independently, tracks the state of the job, and stops when something needs human attention.
@gregisenberg The inbox won’t disappear. It’ll become impossible for humans to consume directly.
That’s part of why we’re building Demux: turn the chaos hitting your inbox into the interface you actually need.
7/
One inbox could therefore become:
a client request tracker
an invoice tracker
a shipment tracker
an approval board
a sales pipeline
an operations dashboard
You decide what you need.
6/
So instead of repeatedly reconstructing what’s happening from your inbox, you can see the current state in seconds.
And you can create different views for different needs.
You receive so many emails that it’s hard to keep track of everything.
We are building Demux to solve this.
Tell it what to track and how to see it. It builds a custom view that stays updated as new emails arrive.
Reserve early access: 50% off your first year.
https://t.co/xo2SxElajH
@anthonylazkani2
AI needs something like the IAEA.
An independent international body that continuously tracks frontier model capabilities, measures how fast the frontier is moving, and raises the alarm when certain thresholds are crossed.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: https://t.co/OGyPb7yaYt
I'm learning that "interesting" and "valuable" are very different words.
We launched https://t.co/xo2SxElajH and we're still figuring out whether people find Demux genuinely useful or just interesting enough to click.
PostHog is making that distinction painfully clear.
42% bounce rate.
People are landing. Some are curious. But curiosity isn't the goal.
The goal is: "I have this problem. I want this."
So now we're obsessing less over getting more traffic and more over understanding why people leave.
Building is easy compared to figuring out what people actually value.