Co-founder & CTO at @Larridin (a16z, Google, Bloomberg backed!) Prev: Retail Trading @ Coinbase
Mountain View
Joined March 2009
- Tweets926
- Following1.9K
- Followers531
- Likes868
Pinned Tweet
This was fun conversation! Thank you for the opportunity!
Scaling tech is easy. Scaling teams isn’t.
@aktwits of @larridin shares how he helped Groupon grow from 37→1000 engineers, what great leadership really looks like, and why AI adoption is more human than hype.
🎧Listen here: na2.hubs.ly/H01zVLv0
#SignalToNoisePodcast
Ameya Kanitkar retweeted
Looking at some @larridin data from September.
The employees getting the most out of AI don't just use it frequently, they use it for many types of work.
Only 20% of employees used AI for five or more types of work. But they delivered nearly 60% of the value and productivity gains from AI.
Those who used AI for five or more kinds of work saved about six hours per week. About 20 times the impact of those who used it for only one kind of work, and saved only 15 minutes.
Even when two people had the same number of AI conversations per week, the ones who used AI for more kinds of work still got more out of it.
For leaders this means you should be pushing your people to continue pushing themselves to continue finding uses for AI.
Share of Vercel deployments made by coding agents:
Jan 2026: 3%
Oct 2026: 60%
Agents are coding. Agents are deploying.
Vercel CEO Guillermo Rauch (@rauchg):
“We started out the year with 3% of deployments on Vercel coming from coding agents.”
“Now, 60% are agentic.”
Also Larridin is proud sponsor!
You need to have a trust system outside of the model providers that you can rely on.
“The models are becoming more and more capable. One of the assumptions I had was that alignment would just fall out of scaling, and it does not seem to be.
The best posture, if you want to continue to use these sophisticated models, is to treat these models in a trust architecture.
If you are really going to have agents that are more capable and yet less aligned, then you need to have a trust system that you can rely on.” @satyanadella
Love to hear your thoughts @zicokolter @rohinmshah @simonw @HamelHusain
Ameya Kanitkar retweeted
Just spent the last 7 days on the front lines of the vulnpocalypse fending off an active cyber attack on one of our products.
Some thoughts:
- Software factories are going to be required equipment for everyone. There’s just no way to keep up with the pace of highly automated adversaries without one.
- We were lucky because we had spent the last few months getting our software factory running well. It is nice to have for product development but proved to be absolutely essential this week.
- At the minimum you need to have a coding server in the cloud pre-provisioned with read only access to all your logs and all the services you depend on for observability. With that in place you are a prompt away from reasonably good situational awareness, ongoing monitoring with loops, and ability for coding agents to validate fixes against your actual, observed attack patterns.
- Shared coding server also unlocks multiplayer incident response where everyone can see every session and jump in as needed. Sharing session links that can be forked is much higher bandwidth than re-explaining things.
- You don’t need to be Ramp/Stripe/Uber scale to have a working software factory in the cloud. We are a team of 3. Ours is built with an @opencode server (thanks @thdxr!), a forked version of compound engineering optimized for automated workflows (thanks @kieranklaassen!) and a multiplayer agent in slack that handles crons (we have our own harness but pick your favorite here).
- To get our software factory up and running well, we had to stop all new product development for ~6 weeks and just focus on getting the validation systems working that enabled us to reliably shipping PR without human code review. This felt very painful in the moment but was totally worth the eventual speedup in velocity.
- KYC for @AnthropicAI & @OpenAI cyber programs is now easy and quick (at least for defensive access), but a day felt like a lifetime this week. Submit your app now before you really need it.
- Getting Mythos access unlocked a deeper level of analysis that helped us shift from reactive mitigation to proactive patching. Wish we had done this a few weeks earlier. Get started now!
Ameya Kanitkar retweeted
Running three layers of leadership at @Oracle does not get an engineering leader off the hook for writing code.
Our @aktwits interviewed Abhishek Agrawal of Oracle. He was in the codebase last week.
For leaders navigating AI, staying close to the work is becoming the job itself.
"If you're not running at the pace where the industry is moving, you're going to get left behind," Abhishek told us.
Full episode in comments:
some inconsistent output is expected... but watch out for sustained drop!
Everyone isn’t at their best 100% of the time. Even engineers.
Most engineering dashboards count output, but how often do engineers create more reviews and rework than they deliver in code?
Here is what we saw when we dug into our @Larridin data on Engineering:
39% of Engineers have one week in a two-month period where the reviews and rework they create outweighs the code they deliver.
The good news is only 2% stay in that pattern for 4 or more weeks out of 8.
If you are tracking the right metrics, you can help engineers turn things around more quickly.
Ameya Kanitkar retweeted
Does AI-written code need more fixing? We looked at the pull requests.
At the same size, AI-assisted PRs get reworked more often. But, matched on work type and complexity, the gap almost disappears: 8.3% vs 7.8%.
AI isn’t shipping broken code. It’s being pointed at harder work, and harder work takes a second pass.
We actually have data on this across 15k engineers.
Some of the top spenders are your best and most productive employees. Blanket spend limits limit your BEST people!
Ok so at @cxodev we talk to a lot of people about their AI spend and individual cap here’s what we tell them:
1. Capping your top spenders is counter productive because 99 times out of 100 they are the biggest impact
2. It’s easy to manage top spenders because you can find them and talk to them about what they’re doing and make it more efficient
3. “Budget divided by headcount is easy but not optimal” - @zachdavis
4. The good AI users will be extremely annoyed by this and do unnatural things so no one comes and bothers them about it
5. 👏 you’re 👏 under 👏 using 👏 your 👏 session 👏 data (to tell where tokens are going)
6. Yes, 2027 is the year when every CFO is going to ask where all the ROI from this spend is
7. No, “we cleaned up tech debt” is not enough
8. Again, 80% of your budget leverage is in R&D, elsewhere you’re probably underutilizing AI. Most trailing orgs are really far behind in adoption
9. Go find your spend centers and eval cheaper models. Probably higher impact than a global cap
10. Your true AI talent will leave if they can’t build / work in the future (uncapped)
An alt proposal to caps: use it or lose it. Pull access entirely for anyone that shows no interest in AI, pour that budget into the people who are driving impact.
Focus on the R in ROI not just the I.
True!
everyone is asking us at @cxodev about calculating ROI on AI, here's what we're saying:
1. focus on engineering, it's 80%+ of your spend - if you're looking to improve cost efficiency its the big lever to move (so stop wasting time in other depts)
2 . model tests + evals are your friend. learn to run them for the most expensive parts of your business
3. spend time with your top spenders: they're either a) doing things your full team should do or b) an easy spot to save some $$$
4. you're probably underspending relative to oppty all things considered
5. PR velocity is a good not great measure of eng ROI. you need to tie to actual product intiatives and use
6. if you're not sucking up session data + traces you can't really tell if your team is using AI on the right things
7. everyone is blowing through their 12 month budget in 4 months. you're not alone
8. aim for 2-10x improvements in ops efficiency. 30% is a rounding error.
9. time saved =/= ROI. what are you doing with that time?
10. you should be eliminating whole categories of human toil
11. nobody calculated the roi of email. some of this is simple work transformation
12. idk we still like tokenmaxxing
13. you will spike up and then it will be possible to pull down costs (but not output) - we see it all the time. intelligence is getting cheaper.
14. revenue per fully loaded employee is emerging as the north star guide. most leaders won't say that out loud to their teams.
15. spend capping your top users is dumb. they're the ones most likely to be highest impact if you really look.
16. sorry, but you just don't have that many fable-level problems. stop making it your defaut.
what am i missing?
Ameya Kanitkar retweeted
Larridin’s Engineering AI Impact Report: Q3 2026 is out, and the numbers show a shift that’s moving faster than most leaders realize. AI-written code is becoming the majority, spending is getting more disciplined, and teams are reorganizing around output.
Results include nearly 2x growth in AI use in pull requests; a sharp drop in spend growth per month; and nearly half of teams reorganizing across specified periods. Per-engineer AI spending varies by 10x between high and low-end spend, and AI-native engineers get twice the bang per buck spent on AI vs. low-AI engineers.
If you're an engineer, engineering manager, or influencing AI usage at your company, this report is a must-read.
larridin.com/data
How engineering leaders are changing the org chart for an AI-native SDLC:
Our survey:
• 40.8% moved to product/outcome teams
• 19.4% changed seniority/role mix
• 15.3% added roles in verification, agent platforms and AI enablement
larridin.com/data/engineerin…
Ameya Kanitkar retweeted
Good news! Larridin will be a Platinum sponsor for ELC 2026. The Engineering Leadership Conference (ELC) annual event next week, October 13-14, at Fort Mason Center in San Francisco.
@aktwits will lead the Main Stage keynote on Tuesday, 10/13 from 1:30-2pm, speaking on: The AI-Native Engineering Playbook: A practical playbook for building high-performing AI-native teams.
Attending? Stop by the Larridin booth to learn more about developer productivity products.
sfelc.com/annual2026#partner…
I tried ASD-STE100 and that works really well!
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
Ameya Kanitkar retweeted
Big news from us today! @Larridin is now live and available through @googlecloud!
Larridin is proud to announce that the Larridin platform is now available through @GoogleCloud Marketplace. Advantages:
Simpler procurement and billing
Private offers, agreed with Larridin
Potential use of existing spending commitments
Onboarding support from Larridin
Larridin answers the question every CxO has to be ready to answer: What are we getting from our AI investment? Plus Developer Intelligence and Workflow Intelligence.
Reach out to us to learn more.
Links:
larridin.com/marketplace
Ameya Kanitkar retweeted
Larridin is proud to announce that the Larridin platform is now available through @GoogleCloud Marketplace. Advantages:
Simpler procurement and billing
Private offers, agreed with Larridin
Potential use of existing spending commitments
Onboarding support from Larridin
Larridin answers the question every CxO has to be ready to answer: What are we getting from our AI investment? Plus Developer Intelligence and Workflow Intelligence.
Reach out to us to learn more.
Links:
larridin.com/marketplace
Ameya Kanitkar retweeted
Thanks to @saastr and @jasonlk for featuring Larridin this week.
With budget season here, now is the time to truly understand your AI investment. As Jason puts it, "more budget converts to more output only where fluency already exists."
Larridin gives companies clear visibility into their AI spend, from tokens to adoption, and connects that spend to the business outcomes it drives.
saastr.com/saastr-ai-app-of-…
Ameya Kanitkar retweeted
Great post from @jasonlk on @Larridin and how people are leveraging our tools in multiple areas. Thanks Jason!
saastr.com/saastr-ai-app-of-…
Ameya Kanitkar retweeted
Tokenmaxxing is out. Valuemaxxing is in.
Enterprise AI spend is projected to jump 63% this year. Boards are starting to ask what they're getting for it.
Read more in the new Forbes article featuring our own CTO @aktwits.
He says, "The companies able to continuously match the cost of AI to the value of the task will likely see better ROI for their AI."
forbes.com/councils/forbeste…
