me: how much complexity is this?
codex: Moderate. Iโd budget 2โ4 engineering days
me: ok do it
codex: * worked for 3 minutes *
codex: done
Pieces of computer history lore you should know:
- Worse is Better
- The LISP curse
- Conwayโs Law
- The Blub Paradox
It's funny how on some teams sending AI-generated messages to colleagues is totally fine while in others it's an absolute tabu. And it's within the same organization!
Pieces of computer history lore you should know:
- Worse is Better
- The LISP curse
- Conwayโs Law
- The Blub Paradox
overheard: "โฆ like in old golden FAANG times: 1 PR a day keeps the manager awayโฆ"
Looking at the code produced by models and calling out bs still gives you an edge, but knowing when to look is a fine art form that keeps changing.
It's been never more important to know what's important.
Show more
It's funny how on some teams sending AI-generated messages to colleagues is totally fine while in others it's an absolute tabu. And it's within the same organization!
overheard: "โฆ like in old golden FAANG times: 1 PR a day keeps the manager awayโฆ"
Looking at the code produced by models and calling out bs still gives you an edge, but knowing when to look is a fine art form that keeps changing.
It's been never more important to know what's important.
Show more
Noticing an interesting pattern with cost of new code nearing zero.
In the past building something was hard and expensive, so teams collaborated and coordinated to reuse shareable pieces.
Now a single engineer can copy a really complex system in days, and there's no incentive to collaborate.
On one hand it's a good thing, coordination is really expensive: alignment, planning, interruptions, requirements reconciliation. But on the other hand we end up building fifteen copies of the same thing, and bringing them to play nicely together is even more expensive than before.
Show more
Books worth re-reading:
The Goal, 1984
The Design of Everyday Things, 1988
The Mythical Man-Month, 1975
Thinking a lot about software factories and testing lately. Here's free alpha:
Anything that adds more noise to the system is not gonna get traction.
Your agent can find bugs from twitter posts? Cute. But if your product is popular enough to have people on X complaining about it, your team is already overloaded with feedback, bug reports and XFN asks. Your twitter bug reporter bot will drawn in noise. So you make your bot fix the issue and submit a PR. Great, now the PR is getting lost in PR noise and your eng team is already overwhelmed with code review.
Show more
"Weโve got all the building blocks, now we _just_ have to connect them all". The real secret saunce is in connecting.
There's still alpha in working with your agent interactively
In this era of software engineering there's no excuse not to ship.
In the past you would have to shepherd orgs and get onto some teams H2 roadmap, but today a motivated IC can simply get stuff done.
Software Factory is a function of your organization's culture.
That's why it's so successful for solo builders, it's O(1)!
I was whispering to my Codex using dictation the other day and my kids thought I speak parseltongue
Old way of deploying:
$ git push heroku master
New way of deploying:
$ codex /goal "keep my git:โฆ app deployed and healthy"
Documentation is officially dead.
Code and Slack are the source of truth (have always been). Codex can reconstruct a "view" into the system on demand, and tailor it to the viewer.
No point in maintaining docs that will become stale the moment they are published.
Show more
Context is everything
