pjm331

Listen friend I’m also in the camp of LLMs are just math and data and don’t actually “have an idea” about anything but that doesn’t make them any less accurate/effective at generating tokens

I’m not sure why this prompt is different than any other in that regard

Yeah i have a similar setup at the harness level - a “devlog” at the end of every session about the experience developing, what could have been better, what was confusing

And also files issues for blockers

I’ve absolutely caught things and made improvements just from skimming them occasionally - they are particularly useful when you get a PR that makes you scratch your head

But I’m definitely not taking full advantage of all the feedback coming in yet

I have to imagine parsing signal from noise there is a massive challenge when it’s other agents that are using your MCP and not just your own

Nothing teaches you like maintaining something for a decade

great game well done i could easily play this for far too long

i have a vague memory of hunter s thompson talking about sitting down and typing out the great gatsby to see how it would feel to write a great american novel

Yes I had a fun experience where it kept on timing out on a seemingly mundane task and it turned out I had written the ask in a way that was impossible to test

and i `brew update && brew upgrade --greedy` every morning with my first cup of coffee because i like to live on the edge like that

thanks for all your work!

so i think the thing that everyone building these git alternatives is missing is a multi-repo story - unless the expectation is that everyone is going to start operating out of monorepos

i've settled on all of this context attached to issues in a project management system and referenced from commits

it works just fine - its not like your agent cannot read your issue tracker

i feel fairly certain everyone has some set of activities or tasks they feel this way about

my wife and i have two non-overlapping sets haha you can imagine how that plays out

https://pragprog.com/titles/lhelph/functional-web-developmen...

don't let the title fool you - the first half of the book is just elixir

over the past 8 years this is the book i've used to ramp back up on elixir and it works like a charm every time - i've never finished it

for me, a mark of a good programming book in this tutorial-project style is that I have started it half a dozen times and never finished it because at some point before the end I've been equipped w/ the tools to go off and do my own thing

> "But pattern‑matching is not system understanding, and plausibility is not correctness."

Why not? Who says that? Who proved that system understanding is not just more complex pattern matching?

I'm not in the camp of "system understanding is just more complex pattern matching"

but I am absolutely in the camp of "there are many tasks where pattern matching is just as effective as actual understanding"

yeah this is one of the few AI-related products that I have seen that make sense to me

but i also wonder to what extent this needs to be its own thing or if this is just something that it looks like we need but really people just need to shovel more stuff into their data warehouse / data lake that you never had reason to before, because now that's all fodder for agentic search

sounds very familiar to what I ended up doing on my internal system - especially anything to do with search - much better to just sync everything to a DB and give the agent access to the DB

it's straightforward to spin up a custom MCP wrapper around any API with whatever access controls you want

the only time i reach for official MCP is when they offer features that are not available via API - and this annoys me to no end (looking at you Figma, Hex)

don't see it in the comments yet so: https://www.brain.fm/

Yeah I’ve had a lot of success with agentic search against a database.

The way I think of it, the main characteristic of agentic search is just that the agent can execute many types of adhoc queries

It’s not about a file system

As I understood it early RAG systems were all about performing that search for the agent - that’s what makes that approach “non agentic”

But when I have a database that has both embeddings and full text and you can query against both of those things and I let the agent execute whatever types of queries it wants - that’s “agentic search” in my book

the actual argument being made here:

"Assign agents the biggest piece justifiable. I can summarize a product outcome or a feature in two lines. That’s what goes on the ticket. Let the agents figure out subtasks when the work is ready for review, not before. Once you break an initiative into technical issues upfront, the outcome gets lost and the focus shifts to minutiae."

This is not about the ticket being well defined, this is about the agent having the larger context of what you are trying to do

I've had the same thought recently and this definitely is a thing that you can do - but there are also cases where you get dramatically better results if you put some more effort into your setup.

e.g. spend time creating a skill about how to query production logs

What is SCP

Im sure that’s possible but my own working style is too chaotic for saving workspaces like that haha

yeah the CTRL+D definitely gives me problems from time to time but thus far i have been too lazy to fix it

"Software people are not alone in facing complexity. Physics deals with terribly complex objects even at the "fundamental" particle level. The physicist labors on, however, in a firm faith that there are unifying principles to be found, whether in quarks or in unified field theories. Einstein repeatedly argued that there must be simplified explanations of nature, because God is not capricious or arbitrary.

No such faith comforts the software engineer. Much of the complexity he must master is arbitrary complexity, forced without rhyme or reason by the many human institutions and systems to which his interfaces must conform. These differ from interface to interface, and from time to time, not because of necessity but only because they were designed by different people, rather than by God."

- Fred Brooks, No Silver Bullet

here is my entire config

    hs.hotkey.bind({"ctrl"}, "D", function()
      hs.grid.show()
    end)

(1) ctrl-d to see the grid, (2) type the letter where you want the top left corner of your window to be, (3) type the letter where you want the bottom right corner to be

window resized

The closest thing I have found that fits right now is just Linear or [insert your own project management tool with a good api here] and then you manage agents in many of the same ways we’ve been managing human engineers for the past decade - assign them issues

My bet is that the last item is what we’ll end up leaning heavily on - feels like the path of least resistance

Throw in some simulated user interactions in a staging environment with a bunch of agents acting like customers a la StrongDM so you can catch the bugs earlier

as much as you or i may be against it, inevitably AI coding will move away from human review and toward more automated means measuring program correctness

this was already happening even before AI - human review is limited, linting is limited, type checking is limited, automated testing is limited

if all of these things were perfect at catching errors then we would not need tracing and observability of production systems - but they are imperfect and you need that entire spectrum of things from testing to observability to really maintain a system

so if you said - hey I'm going to remove this biased, error prone, imperfect quality control step and just replace it with better monitoring... not that unreasonable!

The sci fi version of the alignment problem is about AI agents having their own motives

The real world alignment problem is humans using AI to do bad stuff

The latter problem is very real

It’s funny because on one end of the spectrum you have non dev vibe coders for whom every log is noise

On the other end are the hardcore user orchestrating a bunch of agents, not sitting there watching one run, so they don’t care about these logs at all

In the middle are the engineers sitting there watching the agent go

I’ve had a fantastic experience building out an internal AI agent service using elixir and phoenix - after only dabbling with it in side projects for almost a decade

OTP fits agents like a glove.

You don’t need to implement every api endpoint as a tool you can just say - this is the aws cli tool it takes one string as an argument and that string is an aws cli command

No difference between that and using the bash tool - except you can keep the keys on the MCP server