Also, every time I select a new guide, some of the elements that do not change are still animating and it's distracting.
We should not clear the current docs or experience a flash of loading state when changing routes... make sure we have the necessary primitives to make this the pit of success path for the future
That flash on docs nav shouldn't happen: launching an agent to audit route transitions app-wide and put the right primitives in the pit of success.
First pass is in: docs keep a persistent page frame, and routes share one loader that keeps the last good content on screen. 20 account/admin routes still hand-roll their loading.
On the loading thing20 account/admin routes still hand-roll their loading.I want those fixed now
Merged: every hand-rolled account/admin route is on one shared loader with last-good content; a lint rule bans the old pattern from routes.
Message Cole
Make data-driven decisions
ColeCoder
What if we introduced a second way to call execute? Instead of providing code you'd provide a module and input... whether this would give us significant savings. My intuition says that it would
Your intuition is directionally right, with numbers. Caveat: ... this is a ceiling + realistic band, not a precise forecast... Recommendation: ship it.
I think we need to measure more to see whether a change to something so fundamental would actually be useful.
Lighter weight instructions are better. Information is better than instructions.
Agree: don't change execute until we can measure thin vs glue.
Message Cole
Understand / design the system
ColeCoder
Is it absolutely necessary that we keep the workers keyed by the user ID? If two users have an agent that has the exact same code for execute, can we not just reuse that worker?
Makes total sense ... You're not missing a secret reason we must key workers by user for isolation.
Also: a worker's identity includes the deploy version, so every platform deploy starts every worker over. Narrowing the key to only what changes behavior fixes that.
To be clear my goal here is primarily to cut my own costs and then hopefully we can pass those cost savings on to users through higher entitlements or reduced pricing.
...this is a pretty impactful change. I'm fine with the agent shipping it and deploying it autonomously but I want it to do rigorous manual testing of the preview deploy.
Shipped: deploys no longer reset workers. Testing showed sharing one worker across users isn't safe yet, so the per-user key stays. User pricing is unchanged for now.
Message Cole
Think about sustainability
KodyChief of Staff
Our current MRR is around $160 but the Cloudflare bill is looking like it's trending toward over $500. ... Is this just growing pains ... or do we have a problem with our unit economics?
Here's a first pass at the unit economics, using Cloudflare's published pricing.
All right I'm actually not going to read your analysis until you can pull actual Cloudflare data.
Pulled the actual bill: about $502. Not long-term sustainable as structured today.
MRR $160
Bill $502
Ok so the solution is to find ways to cut costs, not increase prices. Are the recommended fixes obvious?
Let's have Cole investigate and if we can't make data driven decisions then we need a way to instrument so we can make data driven decisions
Cole's findings: two line items are about 92% of the bill.
Dynamic Workers ~$249Row reads ~$207
Dynamic Workers (~$249): request params were baked into the Worker code, so every call created a new Worker. Passing params separately cut unique Workers per day by 60-70%.
Database row reads (~$207): every list ran a COUNT(*). Now the count is stored instead.
Message Kody
What if I'm wrong?
How I'd Get Into Software in 2026
Product Engineering: a hallmark of history's greatest Software Engineers