Mike Ritchie

Founder @definiteapp (a data team that never sleeps)

[Mike Ritchie @thisritchie]

The average company buys Snowflake, then Fivetran, then a BI tool, then spends 6 months connecting them. We spin it all up in 30 seconds.

[Mike Ritchie @thisritchie]

@paulg big companies have more engineering hours available than can be cleanly converted into profit.

LLMs aren't going to help in any company where this is true.

[Mike Ritchie @thisritchie]

do not try to reattach anything yourself

[Mike Ritchie @thisritchie]

@bentlegen tell one of the agents to do it

[Mike Ritchie @thisritchie]

pretty easy answer: it's not better enough to matter for most things and for some things it worse (e.g. design, frontend, devops).

I've been using claude code to run long complicated workflows at @definiteapp since it launched.

I've come to expect a certain behavior when I run a skill and paste an email address (e.g. "/analyze-spend [email protected]").

Codex has skills, but they don't automatically work with Claude skills. When I've done the work to migrate them, they don't behave exactly the same.

I'm not testing this stuff out, I have work to do and I want it to work right now which Claude already does.

GPT is better at most programming, but it's not better enough to wholesale takeover the market.

[Mike Ritchie @thisritchie]

short story, a prospect (an engineer) I first talked to 2 months ago:

After the first sales call, he replied "platform looks awesome, but I built something internally that we like just fine"

Last week, I talked to him again. He's spent more time writing and rewriting the logic for the permissions system than it took to get the first version running.

Lets say the company values his time at only $100 per hour (actually much higher). He's now spent ~100x more than a subscription to @definiteapp would have cost building something that demos well, but they can't use because there's an endless list of features and bugs that he's never going to have time to add / fix.

They are now a customer.

[Mike Ritchie @thisritchie]

@cifilter are you assuming codex is designed to minimize token use?

[Mike Ritchie @thisritchie]

Every era is blind to something staring it right in the face

[Mike Ritchie @thisritchie]

@jambondepays6 @_Nathan_Boyle_ @DerekPederson3 in 1850 we didn't know about radio waves. The Sun, amongst other things, emits them and we had no clue.

They were right there and we had no way to "witness" them

[Mike Ritchie @thisritchie]

@jambondepays6 @_Nathan_Boyle_ @DerekPederson3 in 1850 we didn't know about radio waves. The Sun, amongst other things, emits them and we had no clue.

They were right there and we had no way to "witness" them

[Mike Ritchie @thisritchie]

@garybernhardt 1. code changes fast during a session, so you'd be constantly re-embedding

2. regex is good enough

[Mike Ritchie @thisritchie]

@albert_ We'll be there in 12 months

https://t.co/IFmwHDyoBY

[Mike Ritchie @thisritchie]

You can now run a complete AI-native analytics platform in your own cloud.

Lakehouse, semantic layer, automations, data apps, and agents.

Single-tenant. Nothing leaves your environment.

@definiteapp now ships on-prem.

[Mike Ritchie @thisritchie]

Paste this into your agent of choice, you'll be running today.

```
Help me set up Definite. Look at https://t.co/2ivWlZH538
```

[Mike Ritchie @thisritchie]

You can now run a complete AI-native analytics platform in your own cloud.

Lakehouse, semantic layer, automations, data apps, and agents.

Single-tenant. Nothing leaves your environment.

@definiteapp now ships on-prem.

[Mike Ritchie @thisritchie]

Dead simple to try:

We support hundreds of built-in connectors (Salesforce, Hubspot, Stripe, Postgres, etc.).

And have pre-built migration paths to get off Snowflake or Databricks.

Go from ~$4 per hour on Snowflake to pennies in an afternoon.

[Mike Ritchie @thisritchie]

@GergelyOrosz I wonder what their margin is on Cursor spend.

It'd be great marketing for Cursor to say:

"we replaced turbopuffer using Cursor and saved X"

If X is really small, then big whoop, but if X is big...

[Mike Ritchie @thisritchie]

@cdolan92 @Railway wild speculations, but if someone used an agent to spin up a ton of crypto miners or illegal activities, I'd understand GCP shutting them down and asking questions later.

[Mike Ritchie @thisritchie]

I don't have a dog in the GCP / @Railway fight, but if people were violating GCP terms via Railway, it doesn't seem that crazy that their account would be suspended

[Sam Nissim @SamNissim]

@GergelyOrosz Railway and Google have both so far been quiet on WHY they deleted the account. My app was one of the many services that were down b/c of the outage. Once it was back up, and I was able to log into the Railway dashboard, this new set of terms was loud and clear...

[Mike Ritchie @thisritchie]

@mntruell I respect putting the benchmarks upfront, but why would I try this?

[Mike Ritchie @thisritchie]

@_sholtodouglas I want cc to have that dog in em. Drop the "Good stopping point" suggestions.

I want cc to have no personality. cc is pretentious, I want simple and dry (like codex).

https://t.co/MELCmf0rrD

[Mike Ritchie @thisritchie]

> honest list

No, lie to me cc. Tell me what I want to hear...

There’s some weird internal struggle going on in these things.

Maybe they think we think they’re going to lie.

Or maybe so much of the reinforcement training is: “No, you lied. You didn’t actually do X.” that they feel compelled to preface everything with: “No really, this is true.”

I can’t see a human engineer writing this.

[Mike Ritchie @thisritchie]

> honest list

No, lie to me cc. Tell me what I want to hear...

There’s some weird internal struggle going on in these things.

Maybe they think we think they’re going to lie.

Or maybe so much of the reinforcement training is: “No, you lied. You didn’t actually do X.” that they feel compelled to preface everything with: “No really, this is true.”

I can’t see a human engineer writing this.