Software engineering is phoenixing.
Companies are downsizing.
Agent swarms are promising to replace entire engineering teams.
And yet, the best engineers are busier than ever.
@poteto Chat is too linear. Particularly when combined with async actors. UI that leverages spatial memory for mise en place information combined with attention triage aggregation.
Omarchy being so fast to install opens up a whole new gaming genre: the OS gaming genre. For example, an MMO where losing in the game bricks the device and requires reinstalling the game OS to respawn. Too stupid not to do @dhh@tobi ?
@ThePrimeagen You're right to push back on that, and it is worse than your statement implied.
To state the honest position, I really like signposting my next statement.
@jessitron I've been using a wrapper I created called Swarmforge (https://t.co/TnnV6zWNMf). It supports aliasing commands for separate work/personal usage with different host paths mounted for ~/.claude inside a container.
@housecor > AI can simultaneously review code in multiple dimensions
I have any particular agent review session focus on a specific dimension. Using the dimension definition as a lens to focus its attention mechanism. Then have multiple different lenses run in parallel.
@bentlegen This is the type of behavior I wanted to avoid when I started building https://t.co/iubZqg9TX6
Containerizes the harness. Spins up side car processes based on config (e.g. GitHub MCP that hold credentials) to limit scope of actions agent can take.
@spacexai having @grok in all Tesla's is a strategy similar to Apple's in the 80s-2000s using schools as a distribution channel to influence kids at an early age. Build trust and loyalty reap the benefits in 10-20 years when they release more consumer devices.
Random thought: Enumerating things an agent should not do is potentially dangerous, especially when nearing the decoherence limit of attention.
An ironic rebound effect for agents.
Don't think about `rm -rf /`!
@mitchellh I thought the post was great! I agree with the premise. But I am also the type of person to have books like modern x86 assembly language programming on my shelf.
@paulg Largely agree. The smartest will use LLMs as sparing partners. Find gaps in your own arguments. Persuasion: the derivative of the mind with respect to sensory inputs. Writing rhetoric sharpens your own gradients.
Here is an example of an elliptic curve that has the maximum number of solutions in F_p <= 101 besides F_2
```sage
E = EllipticCurve([110997327521709493305623994548879411633, 86405391492058168448083591898668285015])
print([E.ap(p) for p in primes(102)])
```