Oreoluwa

dev. founder. investor. building software for tedious work: books that won't reconcile, calendars that overshare, events that won't plan themselves.

[Oreoluwa @__reico]

@mfussell gateways feel like the firewall phase, once agents carry their own identity they shrink to a policy check. the byoc thing is the bigger story to me tho. if it all runs in the customer's cloud, what does durable execution even charge for?

[Oreoluwa @__reico]

@bspector @xai thinking out loud: the models agreeing on passing tests can both be wrong about the spec. who wrote the tests, one of them or you?

[Oreoluwa @__reico]

@bspector @xai curious what breaks the tie when both are green but they still disagree 😂

[Oreoluwa @__reico]

hot take: if you're a solo dev building right now, the cofounder you're missing probably isn't technical anymore.
it's the person who already knows the buyers. code got cheap. getting 10 people to pick up the phone didn't

[Oreoluwa @__reico]

@JamesMalsawm neither, it's just following the money. enterprises buy where their cloud credits already sit. hosting a rival beats watching that spend leave for another cloud

[Oreoluwa @__reico]

@TTrimoreau i think mostly yes, but the launch day isn't the waste, expecting users from it is. the real value is the backlink and a page that ranks for your name for years. treat it as seo, not a launch

[Oreoluwa @__reico]

@AlexBurlis wow! congrats. ooc: what was the first sign it was taking off, signups or someone you didn't know sharing it?

[Oreoluwa @__reico]

@skpnky weirdly, i keep going as long as i still use it myself. most of mine started as tools for me, so "quit" only happens when something better replaces it

[Oreoluwa @__reico]

@aymanalabdul @marclou @denk_tweets the early people forgive bugs because they want to shape it. the trap is building only for their requests and never for the person who just wants it to work

[Oreoluwa @__reico]

spent way too long on a bug where every event time was 4 hours off
account had no time zone set so it quietly fell back to UTC
fix was boring: ask for the time zone at signup. should've been there day one

[Oreoluwa @__reico]

@himanshustwts tool-call parsing is the part that quietly breaks on every model swap. same prompt, one model gives clean json, the next wraps it in markdown or invents a field. nice to see someone open source that layer

[Oreoluwa @__reico]

@willia_mbrown and the first customer teaches you fast that the feature you're proudest of isn't why they paid. usually it's some boring thing

[Oreoluwa @__reico]

A confirm dialog asks users to be sure before they know what will happen.
An undo button lets them find out and change their mind.
Most "Are you sure?" prompts are really a team that didn't build undo.

[Oreoluwa @__reico]

Some logic are too tangled to review by reading it, like a state sales tax return built from one giant query.

Where I've landed: don't review the logic, test the result. Fill the form from source data, then diff it line by line against a copy you know is right.

[Oreoluwa @__reico]

Building got cheap. Being trusted didn't.

You can ship an app in a weekend now. Getting one person to rely on it on Monday still takes months of showing up and not breaking.

[Oreoluwa @__reico]

If a SaaS can end up sold for its data, its customers should be able to leave with that data first.

Export isn't a settings page. It's a promise about who that data belongs to.

[@levelsio @levelsio]

Data brokers are now buying up SaaS data sets for anywhere from $100K to $2M

SaaS are struggling now and some extra money to fill the gap by declining revenues is nice

That data is then re-sold to AI training companies and AI frontier labs for the exact purpose of integrating whatever that SaaS did into the AI frontier lab's app

Pretty obvious to see where this is going

[Oreoluwa @__reico]

Background agents move the bottleneck from doing the work to reading it.
100x more runs means 100x more output nobody has checked. I think the products that win eventually will make review cheap: small changes, the evidence attached, and a reason for every edit.

[Aaron Levie @levie]

Moving from chatting back and forth with agents, to having agents that you spin up in the background to do useful work for you is probably on the order of 100X or 1000X more token consumption than what we’re seeing today.

Most work right now is essentially rate limited by our ability to process and work with information. Moving data between systems. Reviewing a document. Looking through code for security issues. Processing log data to look for even patterns. Looking up information about a prospective client. Almost all of knowledge work is constrained by these steps, and AI has become a huge unlock to that.

So our first step was to use AI to do these individual tasks much faster for us. That was the chatbot phenomenon that we have been experiencing for the past 3+ years. But we can still then only augment the work that we remain in the loop for, and we’re highly limited by our ability to parallelize this work by our own bandwidth and context.

But where we’re going is quite different. As AI models continue to improve at incredible rates, we can hand off more and more work to the agent. First it’s letting the agent run with a harder problem. Then it’s giving a few agents multiple parts of a task at once. But we’re not far off from being able to throw an arbitrary number of agents at a problem and just say “come back when you’ve accomplished this goal”.

Now imagine that agent swarm for any problem we can think of. Protecting our codebases, data, and systems. QAing every change we make and make sure all our infrastructure is robust and does what it should. Looking through every document for potential risks. Auditing this company with vastly more data reviewed. And on and on. Most of this isn’t even replacing existing work we do, it’s all net new.

These agents will get deployed as swarms by people, by workflows, by triggers, and by other agents. And none of this would be possible if the models weren’t getting better, and if the cost of AI didn’t make it affordable. Both of those trends though are playing out. Now the next frontier is just having enough compute. But this is where it’s heading.

[Oreoluwa @__reico]

If your agent can edit the tests that declare it done, green is not a finish line.

Done is a user flow that still works when the model isn't grading itself.

[Oreoluwa @__reico]

@paulg I think the advice holds. What changed is how fast you can ignore it.
You used to need months to build the wrong thing. Now it's a weekend. Same mistake, just cheaper to repeat.

[Oreoluwa @__reico]

Sales to agents won't care about your charm. They'll care about uptime, audit logs, and whether the price per task is clear.
If your product only works when a human is patient with it, you're building for a shrinking buyer.

[Paul Graham @paulg]

Did office hours yesterday with a very smart but nerdy founder. When I suggested he might want a more outgoing cofounder to do sales, he said he didn't, because most of his sales would be to agents. I had to admit he could be right.