Prashant Mahajan (@prashantx31) header
Prashant Mahajan (@prashantx31) profile picture

A Product Guy. Leading AI at @monacoGTM, previously built @zedaIo.

San Francisco, CA
Joined April 2014
Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)Photo by Prashant Mahajan (@prashantx31)
We should have more such experiments! Great work @mrsnhl
Humans don't reason from scratch every time we do something. We don't calculate 2+2 each time, or consciously plan every movement when picking up a red cup from the kitchen. We rely on memory, learned habits, and things we've done thousands of times before It's still unclear right now what the equivalent of this should look like for AI agents. How much should they learn through their harness, without changing the model itself? @ArkidMitra and I (mostly him honestly) ran a pre-registered experiment with a frozen 9B model, where a larger teacher rewrote the agent's notes based on its mistakes. We compared that against a teacher doing the same thing without ever seeing those mistakes A few interesting findings for TL;DR's sake: • ~60% of the improvement on isolated decisions came from searching for better notes alone, without needing failure traces • Failure traces added another +4.0 percentage points, but we couldn't demonstrate improvement on full multi-turn conversations or held-out tool-use tasks • A blind audit found that 146 out of 240 held-out BFCL test instances (~61%) had scoring defects. Which raises another question about what we're actually optimizing for The most interesting thing is that making an agent better at individual decisions doesn't necessarily make it better at completing an entire task Over the next few weeks, we're publishing more experiments on what should live in the harness vs. memory vs. model weights, and how to keep agents from breaking as they learn across 3,000+ conversations Paper Link: arxiv.org/abs/2611.11342
2
231
Prashant Mahajan retweeted
I'm excited to share that I've joined @AnthropicAI to lead Startup Partnerships in India! It's super exciting to join a company whose product I use daily and genuinely love. @claudeai has become a regular part of how I think, explore ideas, solve problems and get things done.
411
70
11
3,448
122,727
X is talking about Jensen’s 3rd affair and LinkedIn got to know about 2nd.
8
521
I think these two secretly grab a drink!
3
294
Prashant Mahajan retweeted
God bless the USA (this is a real picture)
18
16
5
177
156,581
Tell me you have not traveled the world, in simple words.
sf has the most beautiful cafes in the world
1
7
539
Prashant Mahajan retweeted
Dear Diwali, Thank you for being the reason I dress up, eat good food, and hang out with my favourite people. 🪔🫂✨
2
1
21
639
Prashant Mahajan retweeted
i think we’re going to see a pretty radical change in how software companies are staffed over the next 3 years.. for the last 20 years, the most expensive thing in a software company was building the software. so we built entire organizations around engineering capacity. PMs figure out what to build. designers design it. engineers build it. sales sells it. CS handles the customer. AI is going to completely upend this.. 1/ the distinction between PM and engineer will start to disappear. PMs who can’t build and engineers who don’t understand customers will both struggle. 2/ the best engineers will spend a lot more time with customers. understanding the problem, building the solution, seeing if it actually works.. without 5 layers of handoffs. 3/ the value of customer understanding will go up dramatically. not because AI can’t talk to customers, but because building 10 features is becoming easier than figuring out which 1 actually matters. 4/ distribution becomes even more important. when 100 companies can build roughly the same thing, the one that knows how to reach customers and earn their trust wins. my bet is that the best software companies in 3 years will look very different.. far fewer handoffs, much smaller teams, and people who own problems end to end. we spent 20 years organizing software companies around the scarcity of engineering. that scarcity is going away.. but the org charts haven’t caught up.
13
9
8
127
18,445
Building agents for support, finance, healthcare, or sales? Nov 18, SF: the Vertical Agent Meetup by Monaco. Evals, memory & deployment talks. Food, drinks, and folks from Thinking Machines, Applied Compute, Braintrust & Judgment Labs. 6:30-9pm. RSVP luma.com/monaco-cdlb
1
5
417
Prashant Mahajan retweeted
For one night, we're bringing together the best agent builders in the Bay to talk shop. On Nov 18, Monaco is hosting the 𝗩𝗲𝗿𝘁𝗶𝗰𝗮𝗹 𝗔𝗴𝗲𝗻𝘁 𝗠𝗲𝗲𝘁𝘂𝗽 in San Francisco. The future of agents is specialized systems for customer support, finance, healthcare, and sales, and this is for the people shipping them in production. Learn and share the practices that work. All signal, no hype. What we'll dig into: 𝗘𝘃𝗮𝗹𝘀: how to know your agent works and catch regressions before your customers do 𝗛𝗮𝗿𝗻𝗲𝘀𝘀𝗲𝘀 𝗮𝗻𝗱 𝗺𝗲𝗺𝗼𝗿𝘆: tool design, context engineering, what an agent should remember and forget 𝗗𝗮𝘁𝗮 𝗮𝗻𝗱 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁: where post-training data comes from, and what breaks when agents meet real customers We'll have expert talks, plenty of food and drinks, and attendees from the best agent companies on the planet including @thinkymachines, @appliedcompute, @braintrust, @JudgmentLabs, and many more. 𝟲:𝟯𝟬 𝘁𝗼 𝟵𝗽𝗺.
5
6
1
27
7,193
Honored to be nominated for Cosign Product Leader 2026 along with some big names. cosign.co/awards/product
ok big drop today: very excited to announce The Cosign Awards 2026 has been a crazy year. One thing we keep seeing: the people making the biggest impact inside top companies aren’t always the people you’d expect. The org chart no longer tells the whole story. a single, standout engineer, designer, growth leader, or operator can change the trajectory of the entire company. If you’ve worked with someone like this, you know... the rest of the world hasn’t caught up yet. We want to help change that. Today, we’re opening nominations for the @Cosign Awards. For the next month, we’re looking for the most impactful contributors, and rising stars across the key functions that make great companies work. We'll aggregate nominations across 41 categories, and with help from a panel of 21 judges including @lennysan, @omooretweets, @packyM, @gokulr, @rrhoover, @dhaber, @avichal, @gaganbiyani, @amasad, @illscience, @venturetwins & more - we’ll select ten people in each category, and then bring them together to meet, exchange ideas, and build relationships. (this is the part i'm most excited about, to be honest .. gathering the winners IRL + group chats: we'll run this every year, and hope to see alumni communities of awards winners building things together, helping each-other out, and more - over time). So, here it goes! Who comes to mind? Nominate at cosign.co/awards
5
1
21
617
Prashant Mahajan retweeted
Excited for the @MonacoGTM poker tournament. Thanks @samdblond for the invite!
1
2
15
7,980
Looking to talk to sales folks who are using Claude/Grok Bot/Codex or any other AI agent workflow to create and close pipeline?
4
10
512
Prashant Mahajan retweeted
Our Agent Platform replaces point solutions and solidifies Monaco as the only sales platform you’ll ever need. It allows you to create agents that can work on your instance 24/7, alongside the ones that we can ship to your instance on day 1. There is no limit with what you can do with these agents and we are here to help you craft best agents that would increase your ROI of Monaco. I’m excited to see the agents our customers will build! Shoutout to @mihail_eric, @prashantx31, Yunshu Guo Adam Tazi, Kulvinder Lotay for being the masterminds of our agent platform. Such a great work 🙏. PS: Our internal agent hackathon produced tens of agents, and some of them are shipping in today’s release.
Today Monaco is launching Agent Matrix. Monaco unleashes countless agents working together, goal seeking, with one outcome in mind: accelerate revenue growth. If you don’t have an army of agents working 24/7 to accelerate revenue growth, you’re living in the past. Apply for early access now.
1
2
14
744
It has been fun building Agent platform at @MonacoGTM. Would love to have more feedback.
Today Monaco is launching Agent Matrix. Monaco unleashes countless agents working together, goal seeking, with one outcome in mind: accelerate revenue growth. If you don’t have an army of agents working 24/7 to accelerate revenue growth, you’re living in the past. Apply for early access now.
3
15
551
Prashant Mahajan retweeted
100+ of SF’s most cracked GTM engineers follow everyone at gtmstack.fyi who did i miss?
98
17
62
216
581,122
We are starting a new trend.
well done @MonacoGTM excited to see what trend this kicks off in sf
1
1
12
801
Prashant Mahajan retweeted
Replying to @prashantx31
well deserved dude!
1
1
66