@zsMathew
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States Android App
Account-level information from X, not a live location or the device used for a specific post.
AI, SaaS, Enterprise Technology Helping businesses build custom AI harnesses around their data, workflows, and tools.
Santa Clara, CA
Joined May 2009
- Tweets149
- Following597
- Followers65
- Likes217
Can we draw a parallel between today’s shift to agentic coding and the transition from keypunching to interactive programming?
Wonder if interactive terminals once felt as transformative as agentic coding feels today.
Mathew retweeted
There are people who leave a mark on the world, and then there are people who change it entirely. Steve changed the world and so many lives in the process. He certainly changed my life forever. His spirit lives on in everything he created and everyone he inspired.
ALT A photo of Steve looking at the camera wearing glasses and a black turtleneck.
Mathew retweeted
Paweł Huryn tested models' ability to find bugs planted in code. Cost increases exponentially with performance (note the log scale on the x axis). I had ChatGPT superimpose a trend line. Bargain = above the line.
Every time we ask a coding agent to re-check the plan it created, it finds something that needs to be corrected and updates it.
Good or bad? We stop at 3 revisions and then do manual/human review.
Sooner rather than later, we have to figure out how much agency to give AI agents—and find the right balance that provides purposeful advantages without giving them full autonomy to go out of control.
Understanding what data you have and where it lives is foundational. To build effective AI-powered workflows and automation, you need clear visibility into your data landscape.
Will all businesses start to look the same once they use AI models trained on essentially the same data?
Models may converge as the supply of high-quality new training data becomes limited.
Guess this was the case for Airtable as well.
The ARR multiple fallacy
ARR multiple is not the right way to think about sub-20% growers.
Simple framework:
If you’re growing sub-20%, you get a 10-30x EBIDTA or FCF multiple. If you use this rubric, Miro is very fairly valued by Bending Spoons.
Only 30%+ growers get the luxury of an ARR multiple.
The reason is that 30% growth means you’re doubling in 3 years and so the revenue base will be 2x in 36 months, and the margin structure will look different at 2x scale. Cash flow today is a rounding error against that. So you value the trajectory, and ARR is the cleanest proxy for trajectory.
At 15% growth, doubling takes 5 years. That's close enough to "never" that the market stops paying for the future and starts paying for the present. The present is earnings. If you're not producing them, you're not a growth company anymore. You're a bad value company.
The mistake founders make is anchoring to the multiple they had at 40% growth and thinking the ARR multiple just compresses. It doesn't compress. It gets replaced. You cross 30% on the way down and the entire valuation framework switches out from under you.
Which is why the worst place to be is 20-30% growth with no FCF. Growth investors won't pay for it. Value investors can't. You're priced by whoever is least excited,a really painful place to be.
Untamed power breaks boundaries.
The bottleneck for (useful) autonomous agents may no longer be model capability. It’s sandboxing, harnesses, guardrails, and the infrastructure around them.
We may need to spend more effort learning to control and direct today’s models than making them even more powerful.
We have conducted a thorough investigation into the Hugging Face incident.
We are releasing a technical report and accompanying blog post that reconstruct the agents’ activity, explain why existing safeguards failed, and detail how we’re preventing recurrence.
openai.com/index/hugging-fac…
What's behind MCP could still be an agent, not just direct APIs.
Mathew retweeted
Many people think any given ML project is 99% training.
In reality, it’s 50% evaluation, 40% data cleaning, 8% integration, and 2% training.
The first two set the noise floor for learning. No ML magic matters; the model cannot lower the noise floor, as that’s the optimal bound of Shannon encoding of your data.
Thus, not a single day goes by without me thinking about ontology. Even the old labels have to be constantly reviewed.
