Chris Dabatos

Husband first. Dad second. Engineer third.

Built - your AI study buddy

[Chris Dabatos @RealChrisSean]

The more I build with agents, the more I want environments that are cheap enough to create casually and durable enough to matter.

Throwaway compute is useful but throwaway compute with memory is where it starts to get interesting for me.

[Chris Dabatos @RealChrisSean]

@TheRohanVarma This is awesome

[Chris Dabatos @RealChrisSean]

@trashh_dev 🤣

[Chris Dabatos @RealChrisSean]

@bansal_io @flydotio I did not know this. Thats generous even if its only 23 cents.

[Chris Dabatos @RealChrisSean]

I don’t think agents want dashboards first. They want workbenches.

Somewhere to try things, leave files behind, pick up context, break stuff, and eventually ship the useful part.

[Chris Dabatos @RealChrisSean]

https://t.co/b4D8JWa5Ep

[Chris Dabatos @RealChrisSean]

"ai writing your code" demos are less interesting to me now. i think the interesting part is what happens after.

can it run the app, debug, persist changes, recover from mistakes, and leave me with something I can actually deploy?

[Chris Dabatos @RealChrisSean]

@botirkhaltaevv @modal Congrats!

[Chris Dabatos @RealChrisSean]

@yashhq_22 I personally can’t force myself to do anything unless I enioy it. I 100% build because it’s fun as hell.

[Chris Dabatos @RealChrisSean]

@pandeyragini24 your resume definitely helps. but on LinkedIn it’s about the words you use in your profile. are you using keywords that recruiters are searching for? if not, no one will find you. hopefully they do when you apply.

[Chris Dabatos @RealChrisSean]

The more I build with voice AI, the less impressed I am by demos that only focus on sounding clear.

A realistic, clear AI voice is great. But if it mishears me, takes too long to respond, costs too much, forgets context, or doesn’t fit the workflow, it still feels broken and terrible. Personally good voice AI is whether I can actually use it without getting annoyed.

[Chris Dabatos @RealChrisSean]

@Prathkum I agree. Although I do rely on AI to pressure test my decisions too. :)

[Chris Dabatos @RealChrisSean]

Everyone keeps asking if AI is going to replace them. For me it did the opposite. It helped me build RecallMEM, voice agents, and memory systems I didn't think I could even touch yet.

BUT it didn't make me care. It didn't make Postgres or audio less broken. I still had to learn the architecture, I still had to plan it out, and... figure out how not to burn through tokens like an idiot.

[Chris Dabatos @RealChrisSean]

This is nine years of using the same thing until the bill and product limits finally got annoying enough to leave.

And the interesting part is not just cheaper hosting. It’s that the old platform constraints started to become product constraints.

[Zach Wilson @EcZachly]

We're migrating off Heroku after nine years of being a loyal customer!

I deployed the DataExpert Heroku app for the 2541st (and last) time today

2100 of those happened after founding https://t.co/CWvLDHTuVZ three years ago!

Since quitting Airbnb, I've deployed the DataExpert app roughly two times per day every day for three years.

Heroku has been ripping us off. ~$1200/month for performance servers and database.

We migrated to https://t.co/5Af6rhMrkn and now pay ~$175/month for better performance and more flexibility.

Heroku is bad because:
- you cannot apply multiple wildcard SSL certificates onto a single production app. This one problem more than doubled my costs because I needed a separate server for DataExpert and TechExpert.
- the costs on Heroku don't scale nicely. You either pay $50/month for garbage or overpay $250/month for "performance"
- the only pricing tier on Heroku that makes sense is Heroku Postgres which we debated keeping but realized it's better to keep everything within the same VPC on Fly.

These small changes save the business over $10,000/year.

[Chris Dabatos @RealChrisSean]

@EcZachly Huge savings smart decision :)

[Chris Dabatos @RealChrisSean]

@Devin_The_Dev

[Chris Dabatos @RealChrisSean]

Now go install Postgres, pgvector, Ollama, poppler, fix Homebrew, pull 20GB of weights, and explain to yourself why PDFs and uploading images are broken.

[Chris Dabatos @RealChrisSean]

The first thing I learned building RecallMEM is that "private/local AI" is a lie in the funniest way.

The model is local. Great.