Greg Wedow

Task Management for Agents

[Greg Wedow @gregwedow]

Once a programmer becomes used to a complex solution to a problem, simple solutions to the same problem feel incomplete and uncomfortable

— Doug Hoyte

[Greg Wedow @gregwedow]

@htmx_org i really miss discrete buttons

[Greg Wedow @gregwedow]

you know that goes both ways. unless you've tried a modern hypermedia approach, it's probably not possible for you to have an educated opinion on whether it's "completely insufficient" (it's not)

maybe there's a gap, maybe. but i've never seen 10x code to fill a 1% gap turning out to be a worthwhile tradeoff when all that extra code could be filling out the genuinely novel features of a product instead.

[Greg Wedow @gregwedow]

@0xblacklight @dexhorthy sure, building apps on wordpress is a nightmare and wholly insufficient. no one is suggesting that.

modern hypermedia approaches are far more capable than you think. good odds for shipping a better UX in a far simpler package than whatever you're building right now

[Greg Wedow @gregwedow]

@dreamsofcode_io @yinebebt_ It's because embedding the existing tsnet in anything but a Go project requires pulling in the whole Go runtime, which is annoying to interop with from a lot of languages.

They wrote about it here

https://t.co/i803Y5EEAK

[Greg Wedow @gregwedow]

i mean call them as part of scripts. the big unlock with RLM is that `agent()` is available as a function in scripts rather than just giving the model an Agent tool it can call. so then it can programmatically delegate work instead of having to buffer intermediate results within its own context window

[Greg Wedow @gregwedow]

think i need to write a proper full post on RLMs at some point. it's not cheap to run but the results are unmatched

[Greg Wedow @gregwedow]

depends how you leverage it. i'm using the RLM approach as an alternative to compaction. when the context window hits a certain length, the harness moves the entire history into an `archives/` dir and seeds a fresh context with the archives path.

the agent can then script subagents to query the archives and understand what's going on and what needs to happen next. so far that's been surprisingly effective at carrying all important content forward.

since *all* context is in the archives, nothing is ever really lost like when a poor summarization agent drops vital info. the main agent is always able to issue follow-up queries if needed, but often the initial query across the full archive is sufficient.

[Greg Wedow @gregwedow]

depends how you leverage it. i'm using the RLM approach as an alternative to compaction. when the context window hits a certain length, the harness moves the entire history into an `archives/` dir and seeds a fresh context with the archives path.

the agent can then script subagents to query the archives and understand what's going on and what needs to happen next. so far that's been surprisingly effective at carrying all important content forward.

since *all* context is in the archives, nothing is ever really lost like when a poor summarization agent drops vital info. the main agent is always able to issue follow-up queries if needed, but often the initial query across the full archive is sufficient.

[Greg Wedow @gregwedow]

the great religious debate of our time.

one side working from empirical observation, the other holding fast to their dogma.

i could read these all day

[Perry E. Metzger @perrymetzger]

I have said this before, but to those of us using AI systems to get lots of work done reliably and quickly, the people who post online about how AIs still hallucinate constantly, about how they can’t write code, etc., seem equivalent to people trying to convince you that the car you drive to work every day doesn’t exist.

You tell them things like “but I drive a car. I paid money for it. I buy gasoline for it. I could not possibly be working twenty miles away from home if I didn’t have the car?” and they reply that you are imagining having a car, or that you’re lying because you work for a car company.

It is as though these people live in a completely different reality.

[Greg Wedow @gregwedow]

@59thProfile are you doing https://t.co/zzydWsid7q

[htmx.org / CEO of #Schizophreniamaxxing (same) @htmx_org]

https://t.co/a6kTxZtpf3

[Greg Wedow @gregwedow]

@Pensive_dr @NeuralYogi @yacineMTB do we need a formula to tune this? thought the idea here was to just sweep literally everything

computer runs sweeps faster than you can come up with tuning ideas

[Greg Wedow @gregwedow]

@DelaneyGillilan What a stacked lineup. Do you know if it's being recorded/streamed?

[Greg Wedow @gregwedow]

@mark_l_watson Hi Mark, what build of ollama are you running that supports this model? Mine doesn't want to run it.

[Greg Wedow @gregwedow]

@DelaneyGillilan so so so so so many reasons to not be using sync engines these days

[Greg Wedow @gregwedow]

@jarredsumner @yacineMTB have it manage a durable dependency graph of tasks

https://t.co/SN8rtHXsBF

[Andy Gayton @cablelounger]

Quality in the Age of Slop

extraordinary post by @sinclairtarget (via @anders_murphy)

https://t.co/sWOoC5buAp

[Thorsten Ball @thorstenball]

An internal Slack post I shared with the Amp team a few weeks ago:

Building software is learning.

https://t.co/gasUel9aAv

[antirez @antirez]

Many of you forgot too fast the insane amount of shitty software we had to see and suffer in the pre-AI era.

[dex @dexhorthy]

using AI for coding is a deeply technical engineering craft

most people don't approach it as so, and don't get the results we associate with high craft

but the ones who do have been sprinting ahead

more tokens wont save you, more thinking + skill + llm intuition will

have been saying this for almost 9 months now