Andrew Clough

Thank you for posting things like this. I am now doing (conservatively) 2 weeks of old work each day. There is room for improvement.

I say this to non tech people and it's like it doesn't register at all.

Same. I feel my brain decaying while I think about the purpose, structure and impact of code I create rather than the syntax.

That's how I feel about a lot of the Dems, Schumer et al. Thanks for your service, but you're not going to be able to take us where we need to go.

Agree that it makes no sense to vibe code a stripe replacement. I think it's going to commodify these processes.

The emphasis is going to be "Do this one thing effectively and cheaply, then give me my data." There's no need for UI, no platform...

Thoughts? Trying to get my head around it all...

Saw a massive train display built by one guy who had a family and full time job and all I could think was, "Man, they reeeally did not know how to talk about autism in the 50s...."

My take away is anyone who thinks they know what's going to happen, or where the end of these knock-on effects are going to take us, is full of it.

Think it's correlation not causation. On the flip side, this week I've seen:

A LLM prompt that will wipe out 90% of the transform layer in ELT

An oss that makes GTM and segment/mixpanel obsolete.

An sqlite/api program for LLM email search that makes Gmail UI useless.

I feel like over the past 3 months alone I've seen more projects opening up. Maybe not enterprise oss yet, but AI feels like it's giving high quality programmers room to flex on projects they would not have done before due to time/effort.

Put it in the Louvre.

That's where I started.

The first step up is to use Claude code or codex (chatgpt Claude code). It will see your code base and write python directly to your script. Much faster and better context.

Do that for a week, then look into declarative prompts and skills. A lot to wrap your head around!

How are you set up?

I've started two agent flows, one for architecture, one for data expansion (new sources, or new schema on existing sources).

Can't tell if I'm over solving.

There are a lot of examples of software development with agent swarms, but not analytics!

I still have not gotten to the no IDE jump. What's the work/iteration structure that allowed you to do that?

it's like deskilling a hand crank at a factory....

I do data analytics and I've reworked my task processing 3 times in the last 4 months. I don't know when all the knock-on effects will stop rippling, but the starting gun has been fired.

All jobs we have are really just a list of tasks to be done. The only way a job gets replaced is if it wholesale does all the tasks. Jobs will transform around the tasks that can't be done by AI.

one of the unforeseeable benefits of AI coding is that it's quietly stripping away the over-constructed abstractions around the movement and organization of data and information.

Dropbox and Google Drive are designed to serve the companies that created it, not the end consumer.

all my tech issues with Claude code went away when I moved my environment to linux.

add it to the list

Are there people or projects to follow and track for this?

I get that big players don't want this since it commoditizes of all the models. But a Claude code style integration and interface that switches back and forth between complex and simple models for tasks seems like the natural evolution.

This is what keeps catching me. I use Claude, but the vast majority of what I do does not require a high end model.

I'm wondering when there will be a solution like Claude code but I can just switch out for local or cheaper models. I guess Cursor kind of. Any thoughts on this would be appreciated!

Framing this post as a reminder. I build data pipelines and analytics. A real evolution in my work was realizing what I deliver to clients is just a code package that translates data logic into business logic.

This really helped me understand how to understand some of your work better and how I can use it. Thanks!!

@olufemiotaiwo.bsky.social I don't regret to inform you...

Not if you crash everything then start it back up!!!

Thoughts on andor?

Thanks. Was seeing analysis that was saying meta essentially could walk away and leave the banks (and by implication, all of us) holding the bag.

This makes it much more clear that a bubble pop is going to hurt everyone, meta included and that they are hand waving that fact right now.

Plain english this for me.

Basically it's Houdini accounting, but what both of you are saying is that no matter how it's structured, it's hard not to assume Meta is on the hook for the debt?

My bet for the future: AI models are a commodity. While cutting edge will be important, context and an ok model gets you 90% of the way there.

The Chinese models are smaller and open source. Meaning they can be hosted cheaply and data isn't exposed. That's exactly what most companies need.

I feel a Past Lives / buzzfeed quiz cross over potential with this one.

"I apologize for such a long letter - I didn't have time to write a short one."

A quote that lives rent free in my head for all of my work.