Engineering leader building great teams and products @traversal_ai. Previously @wherobots, and @splunk/@signalfx. Tweets and opinions are my own. 🇫🇷🇺🇸
San Francisco, CA
Joined November 2013
- Tweets1.5K
- Following752
- Followers454
- Likes3.2K
The little words of affirmation on the pull-down-refresh of @SlackHQ's mobile app is such a nice touch ❤️ And everything I needed this morning.
It's another day that ends in day. @github and @claudeai are degraded. We're building @traversal_ai so this stops happening. Until then, #hugops to the teams there keeping the most used tools of 2026 going for all of us.
Maxime Petazzoni retweeted
How do you rigorously evaluate AI SRE across both controlled experiments and the complexity of real enterprise production?
Last week, we released ORCA-Bench: a public testbed with synthetic incidents in a realistic simulated production environment. Even there, out-of-the-box frontier models struggled with production RCA.
Today, we’re sharing Traversal-Bench, where we took the evaluation into real enterprise production: actual Sev 1–3 incidents from a large Fortune 100 financial services enterprise, testing Traversal and frontier models with direct access to production systems head-to-head.
The result: Traversal delivered better RCAs, 2–3× faster – and the gap is widening as frontier models improve.
Full article here: traversal.com/blog/the-model…
More on the three key findings below ↓
Maxime Petazzoni retweeted
This was our strongest quarter yet at @traversal_ai !
When one customer experienced Traversal for the first time, they called it the “Claude Code moment for SRE.” Traversal spread organically across their team, unlocking use cases faster than we could sell them. They expanded to more users and use cases within months of going live.
The results followed:
– ARR has grown 8× in the last six months
– Monthly active users up ~14× since January
– Customers ran 71,400+ investigations last quarter, supporting roughly 2,500+ engineers
It all traces back to one early decision: build Traversal around incident root cause analysis for complex enterprises – the hardest reasoning problem in software reliability.
Petabytes of telemetry, almost no labeled MELT data, APIs never built for agents reasoning across thousands of hypotheses in parallel. Solving RCA forced us to rethink both the data and AI architecture.
Customers first trusted Traversal with Incident RCA. Then it spread.
More than half of usage now comes from workflows we never built for – CMDB validation, release validation, data ops, infra resource management, cost optimization.
In one customer-built workflow, an engineer used Traversal to find redundant logging and uncovered $1M+/year in potential observability savings.
This reinforced something we’ve long believed: once AI deeply understands a production system, troubleshooting is just one application among many. The same reasoning foundation extends much further.
Grateful to our customers for pushing Traversal into new corners of their organizations, and to the team for pushing the technology forward.
Maxime Petazzoni retweeted
Coding agents have benefited from a wave of thoughtful public benchmarks over the last few years.
Production AI deserves the same level of scientific rigor.
This week, @traversal_ai 's AI Research team released ORCA-bench, a new public benchmark built in collaboration with @cornell_tech and @Columbia for evaluating AI agents on production root cause analysis.
Paper, benchmark, dataset, and leaderboard: lnkd.in/grr9aRxx
More on why this matters below. 🧵
Maxime Petazzoni retweeted
Today, we are introducing Traversal Workers. No tag required.
The industry's first AI agents built to proactively investigate wherever they're needed, from alert triage and deployment monitoring to complex production incidents, without being scheduled or summoned. It feels like having your favorite SRE teammate seamlessly problem solving with you.
Because AI in production shouldn't just know how to investigate. It should know when to act.
Learn more: traversal.com/blog/ai-sre-pr…
Wait, the company that got the no-bid contract for the reflecting pool is called Greenwater Services? And we're surprised it turned green again? This is Four Seasons Landscaping levels or irony 😂 nytimes.com/2026/06/18/us/po…
Maxime Petazzoni retweeted
how to be good at your job
- realize this one thing is actually made up of two separate things
- realize instead of solving the direct problem you can solve a broader problem
- instead of implementing thing, implement other thing that makes it easier to implement thing
Maxime Petazzoni retweeted
What kind of IC work can you do if you are a manager?
Attended two dinners with dozens of senior Bay Area tech managers last week.
Every single one of them said they were expected to be hands-on building with AI these days. But what to work on?
4 categories emerged:
1. Internal efficiency -- anything to make your team more productive, from communication tools to digital brain to team skills.
2. Quality-of-life improvement -- got something that bugs the heck out of you about the product but you can't convince anyone else to work on it? Just fix it yourself.
3. Celebration story -- artifact (usually video, image narrative, etc) that hypes up the work done by your team.
4. Vision piece -- describes a really freakin' cool future that your team can move towards.
What NOT to do: take on any critical-path product work. (Because then you either do a poor job delivering on that, or managing.)
Maxime Petazzoni retweeted
As agents become the biggest users of software, then all software has to be available in a headless fashion. Agents won’t be using your UI, they’ll be talking to your APIs.
So the question becomes what is the business model of software and this headless approach in the future?
Here are a few thoughts on how everything plays out based on what we’re seeing and doing at Box, but also conversation with other platforms.
1) Seats don’t go away for *people*. Seats are still a convenient and efficient way to have a customer use technology predictably for a set of users within a baseline set of usage. The key, though, is that when the customer pays for a seat, it has to come with a set of usage of APIs on behalf of that user that the agent can use on their behalf.
The user will need to be able to interact with their data and the underlying tool via any agent they work with, and an embedded amount of usage will come with the seat. I would imagine most software -Box included- will enable seats to work with their data at a relatively high volume via systems like ChatGPT, Codex, Claude, Gemini, Cursor, Copilot, Perplexity, Factory, Cogniton, et al. quite seamlessly. If you don’t do this, you’re DOA.
2) Agents may have “seats” if they are doing stateful work in the system, but they will be priced very differently than people. Seats (or the equivalent) can make sense when you have an agent that has its own workspace, stores its own data, needs a different set of permissions compared to the user, and so on.
If a company wants this agent to be around for long period of time, that may very well look like another “user” in the system. Openclaw-style agents highlight what this future could look like.
The only issue on pricing here is that one customer could decide to do all their work in 1 agent, and another might split it into 1,000 agents. So pricing like a human seat is nearly impossible and impractical; each company will have a different approach for this as it gets tricky perfectly trying to capture all the value within an agent seat.
3) The dominant pricing for headless use that goes above the seat allotment, or when an agent is firmly acting on their own, will be a consumption model. Many enterprises software platforms have previously operated like this with PaaS options, and agents will look like another machine user of their system.
In some cases the APIs might get priced just as they did previously, but in other cases there may need to be new types of APIs that represent the work an agent would do in one go -more akin to an outcome- instead of a series of API calls. This is especially germane when the headless software also has an agentic use-case embedded within in, such as orchestrating the process within their own system via AI.
Overall the growth of this usage pattern is effectively unbounded as the use-cases for agents operating on data in these systems will dramatically exceed what people do with their data and tools today. Every platform that goes headless (which will be anyone that wants to take advantage of agents) will need to adopt a model like this. Some may fight it initially but it’s an inevitably as there will always be more and more agents outside your platform than people.
Overall, there’s a lot of really interesting changes left to come in software due to headless use of these systems. Early days.
100% agree and to me this makes the massive investments in inference compute infrastructure and datacenters a super strange bet when more than half of all inference may be done on-device.
Hard to predict of course but when hundreds of B$ are in the balance it's a tough one.
This is a glimpse of big changes ahead of us. If you’re betting on big central models you should think twice.
I run the exact same setup (M5 MacBook, qwen3.6-27B, pi, ollama) and while its not as fast or good as one of the big central models, it’s past the line of “cool demo” into “truly useful.” Kind of where the big frontier models were in late 2025.
In ~24 months we might have local models that are fast and good enough for most tasks.
mfw my weekly average hit 5h55 this week juggling a high intensity job and a toddler: 😱😵
The scariest finding in this paper: the subjects couldn't tell it was happening.
UPenn ran this study on 48 healthy adults. One group slept 8 hours. Another slept 6. Another slept 4. For 14 straight days. They tested cognitive performance every 2 hours from 7:30am to 11:30pm.
The 6-hour group's reaction times, working memory, and sustained attention deteriorated on a near-linear curve. By day 14 they were performing at the same level as someone who hadn't slept at all in 48 hours. The 4-hour group hit that threshold by day 6.
Here's the part that should unsettle everyone who thinks they "do fine" on 6 hours: the subjects' self-reported sleepiness flatlined after the first few days. Their brains kept getting worse. Their perception of how impaired they were stopped updating. The cognitive decline was invisible to the person experiencing it.
The researchers found a hard threshold. Any wakefulness beyond 15.84 hours in a day produces cumulative neurobiological cost. That cost compounds every single day you exceed it and does not reset with a weekend of sleeping in.
About 35% of American adults sleep less than 7 hours a night. 40% of those get 6 hours or less. In 1942 that number was 11%. We built an entire professional culture around a sleep schedule that this paper says is functionally equivalent to pulling consecutive all-nighters.
"I'm fine on 6 hours" is the most common response to sleep research. The first thing chronic sleep debt destroys is your ability to notice chronic sleep debt.
Parenting is like a never ending Easter egg hunt, but with cheerios.
I think my MTBC (mean time between cheerios) has averaged under 15min today. They just keep popping up everywhere 😆
As OpenAI and Anthropic continue to eat up the software development ecosystem, the next step will be for them to acquire and own where+how those applications are hosted.
I'm joining @traversal_ai as Head of Engineering!
We’re excited to share that Traversal has expanded its leadership team with six senior hires across go-to-market and engineering.
Over the past two years, Traversal has moved through three phases: research → validation → production proof.
What began as an applied AI research effort is now running inside some of the most demanding Fortune 100 environments in the world — including Pepsi, Amex, and DigitalOcean — investigating real incidents across petabyte-scale telemetry.
As adoption accelerated, it became clear it was time to scale the team.
Collectively, these leaders bring experience from companies that helped define the observability ecosystem — including Cribl, AppDynamics, Splunk, Google, and Redis.
They’ve seen firsthand how complexity and telemetry are outpacing human operability — and why reliability increasingly requires a rigorous AI-first approach. That conviction is what brought them to Traversal.
We’re excited for this next phase of scaling together.
Read more about the expanded leadership team: bit.ly/traversal-leaders-tx
What F1 needs are the sounds of F3 and the drama of F2. We'll see how the season unfolds but onboards are even more boring to watch than ever with the new regs.
