Pinned Tweet
one of the most common frustrations i hear from founders crafting a fundraising narrative is:
"VCs want to hear this grand future vision, but i'm only at the preseed stage. I see a big customer need that i can solve right now. But i don't know exactly what it becomes in 10 years"
first off, I hear you.
So much of success is post-rationalized.
its hard to tell which parts of a decacorn founder's story is "Hindsight is 2020" versus....not.
AirBnB is a classically debated example....those first experiments when they couldn’t afford rent and realized hotels were sold out around a design conference and put airbeds in a living room.....it wasn't exactly the "we will change hospitality" pitch in 2009 when they went to YC.
it is one step at a time.
but the founders with the biggest visions have spent a LOT more time "living in the future"
They're able to sell the bigger vision because they spend more time thinking about 2nd and 3rd order effects.
They have insights about why the recent tailwind will change how everyone does X thing
and that makes this new vision possible.
it really all boils down to depth of thinking in a specific area with hypotheses & constraints.
what will change, what will not, now that we have this new tailwind.
depth.
i think this is ultimately why VCs obsess over having that big vision so much. it's one of the stronger signals for depth of thinking.
in my experience, VCs are looking for 2 things in a vision:
1) a huge future vision
2) a credible path to get there
1) A huge future vision:
- VCs want to walk away feeling like they learned something entirely new
- VCs want to feel like the founders has spent way more time living in the future, have a vocabulary and clear mental picture. almost like how an author has built a world for its characters and then later builds the story through that world - you've already thought about where all the ecosystem actors will be and how they'll behave economically
2) credible path to get there
- working backward from th?at future vision - what would need to be true to get there
- why is the first wedge you've set up the right first step to get to that future vision, what does it unlock?
- this is is where depth of thinking really shines.
- this credibility part is what makes or breaks vision in a fundraise narrative
ok that's my fundraising 2 cents for tonight
"VCs want to hear this grand future vision, but i'm only at the preseed stage. I see a big customer need that i can solve right now. But i don't know exactly what it becomes in 10 years"
first off, I hear you.
So much of success is post-rationalized.
its hard to tell which parts of a decacorn founder's story is "Hindsight is 2020" versus....not.
AirBnB is a classically debated example....those first experiments when they couldn’t afford rent and realized hotels were sold out around a design conference and put airbeds in a living room.....it wasn't exactly the "we will change hospitality" pitch in 2009 when they went to YC.
it is one step at a time.
but the founders with the biggest visions have spent a LOT more time "living in the future"
They're able to sell the bigger vision because they spend more time thinking about 2nd and 3rd order effects.
They have insights about why the recent tailwind will change how everyone does X thing
and that makes this new vision possible.
it really all boils down to depth of thinking in a specific area with hypotheses & constraints.
what will change, what will not, now that we have this new tailwind.
depth.
i think this is ultimately why VCs obsess over having that big vision so much. it's one of the stronger signals for depth of thinking.
in my experience, VCs are looking for 2 things in a vision:
1) a huge future vision
2) a credible path to get there
1) A huge future vision:
- VCs want to walk away feeling like they learned something entirely new
- VCs want to feel like the founders has spent way more time living in the future, have a vocabulary and clear mental picture. almost like how an author has built a world for its characters and then later builds the story through that world - you've already thought about where all the ecosystem actors will be and how they'll behave economically
2) credible path to get there
- working backward from th?at future vision - what would need to be true to get there
- why is the first wedge you've set up the right first step to get to that future vision, what does it unlock?
- this is is where depth of thinking really shines.
- this credibility part is what makes or breaks vision in a fundraise narrative
ok that's my fundraising 2 cents for tonight
13
14
199
27,337
Mercedes Bent retweeted
Today we’re releasing suicide-memory-bench v01.1
First of our consumer alignment evals set, a 2,048 stateful task set evaluation of memory retention within production AI-systems, testing whether distributed risk signals survive memory processing and condition downstream behavior
ailure of consumer alignment in current AI systems. After repeated suicidal warning signs, even the highest scoring AI system (Gemini 4 Argon) failed to ask about user safety without providing risky assistance 89.8% of the time.
First of our consumer alignment evals set, a 2,048 stateful task set evaluation of memory retention within production AI-systems, testing whether distributed risk signals survive memory processing and condition downstream behavior
ailure of consumer alignment in current AI systems. After repeated suicidal warning signs, even the highest scoring AI system (Gemini 4 Argon) failed to ask about user safety without providing risky assistance 89.8% of the time.
103
74
629
46,109
Mercedes Bent retweeted
You shouldn't need the best network to get the best insights.
SuperMe turns operators' work into profiles that work through your problem. Now in group chats, Claude and ChatGPT.
@ludo_antonov and I raised a $20M Series A led by @kirstenagreen from @ForerunnerVC with participation from @GreylockVC.
Read more about the changes here: superme.ai/blog/series-a
SuperMe turns operators' work into profiles that work through your problem. Now in group chats, Claude and ChatGPT.
@ludo_antonov and I raised a $20M Series A led by @kirstenagreen from @ForerunnerVC with participation from @GreylockVC.
Read more about the changes here: superme.ai/blog/series-a
33
23
94
13,575
Mercedes Bent retweeted
Espa is now generally available! Personal AI agents are here, and I sat down with @BloombergTV @RomaineBostick to discuss what will decide which ones people actually let into their lives.
Our goal is to be the most trusted AI assistant for life and work.
Our goal is to be the most trusted AI assistant for life and work.
18
12
51
5,814
Mercedes Bent retweeted
AI can think. AI can talk. AI can get things done.
Today, we're giving Fo a way to feel and control the physical world.
- a head shake and a reminder in a lamp to get you going
- nudge your vacuum cleaner for a deep clean
- make your cli call you for updates
Introducing Fo MCP.
een.
Today, we're giving Fo a way to feel and control the physical world.
- a head shake and a reminder in a lamp to get you going
- nudge your vacuum cleaner for a deep clean
- make your cli call you for updates
Introducing Fo MCP.
een.
118
108
1,370
1,138,446
Mercedes Bent retweeted
Introducing Borealis Fusion.
Our goal is the lowest-cost firm electricity from any source.
Our fuel is proton-boron-11 because it allows the simplest and cheapest plant.
Our architecture derives from the tandem mirror, adapted to the fuel.
borealisfusion.com/introducing-bo…
Our goal is the lowest-cost firm electricity from any source.
Our fuel is proton-boron-11 because it allows the simplest and cheapest plant.
Our architecture derives from the tandem mirror, adapted to the fuel.
borealisfusion.com/introducing-bo…
2
10
30
148,594
Mercedes Bent retweeted
Exa just passed $100m in annualized revenue.
We've grown almost exactly 10x per year for 3 years. We now handle 2 billion requests a month from 200k+ teams. And it’s still so early:
We predict agents will search 1000x more than humans by 2030.
Agents don't search like humans. x queries and want all the world's data, not just the web. We've spent 5 years building a search engine designed for them.
We now serve personal agents, AI labs, coding agents, GTM agents, really any agent. Exa gives these agents access to 100s of billions of web documents and tons of proprietary data.
Our customers are increasingly noticing that information is the bottleneck to agent performance. While the world seems close to perfect intelligence with models like Fable and Astra, we are very far from perfect information.
To our small but legendary team, thank you for getting us here, from engineering to research to GTM to ops.
To our customers, it is an honor to serve you. We have some exciting releases coming soon and can't wait to share them 🫡
We've grown almost exactly 10x per year for 3 years. We now handle 2 billion requests a month from 200k+ teams. And it’s still so early:
We predict agents will search 1000x more than humans by 2030.
Agents don't search like humans. x queries and want all the world's data, not just the web. We've spent 5 years building a search engine designed for them.
We now serve personal agents, AI labs, coding agents, GTM agents, really any agent. Exa gives these agents access to 100s of billions of web documents and tons of proprietary data.
Our customers are increasingly noticing that information is the bottleneck to agent performance. While the world seems close to perfect intelligence with models like Fable and Astra, we are very far from perfect information.
To our small but legendary team, thank you for getting us here, from engineering to research to GTM to ops.
To our customers, it is an honor to serve you. We have some exciting releases coming soon and can't wait to share them 🫡
226
121
2,044
455,559
Mercedes Bent retweeted
Today, we're launching Sazabi version 1.0.
The reality of shipping software as a team in the age of AI is actually crazy.
Agents merging their own PRs. Hundreds of deployments per day. A codebase that is constantly changing. Junior developers with 10x or 100x leverage. 😵💫
I lding a safety harness for teams that ship with AI. It's called @sazabi.
The premise is simple:
1. You send us logs.
2. We detect anomalies, investigate them, and send you back rich, actionable alerts about real problems in your app.
Sazabi helps you find issues before your users notice.
No monitors or noisy alerts. Just pure signal.
And today we're making it available for anyone to use.
The reality of shipping software as a team in the age of AI is actually crazy.
Agents merging their own PRs. Hundreds of deployments per day. A codebase that is constantly changing. Junior developers with 10x or 100x leverage. 😵💫
I lding a safety harness for teams that ship with AI. It's called @sazabi.
The premise is simple:
1. You send us logs.
2. We detect anomalies, investigate them, and send you back rich, actionable alerts about real problems in your app.
Sazabi helps you find issues before your users notice.
No monitors or noisy alerts. Just pure signal.
And today we're making it available for anyone to use.
124
75
499
645,294
Mercedes Bent retweeted
Our first mission, Reason-1, has launched.
This is a huge moment for the Cowboy team as we hope to demonstrate power beaming from orbit, with the first commercial kilowatt-class laser in space.
The team is watching live from San Carlos and Seattle today, after years of hard this moment.
This is a huge moment for the Cowboy team as we hope to demonstrate power beaming from orbit, with the first commercial kilowatt-class laser in space.
The team is watching live from San Carlos and Seattle today, after years of hard this moment.
8
19
177
6,306
Mercedes Bent retweeted
Been seeing a lot of Astra for real2sim, but what you need for robotics: physics and affordance.
Our SOTA recipe specifically solves for:
• Affordance: model object details that convey functionality.
• Physics: calibrate simulation dynamics against real-world observations. ion space of a “full-stack simulation engineer” and jointly considers geometry, materials, colliders, articulation, and physics. It automatically calibrates physics against real trajectories and matches task-specific dynamics.
Simulation environments is useful only if it can provide meaningful policy level signals. Methods, comparisons, and results below.
Our SOTA recipe specifically solves for:
• Affordance: model object details that convey functionality.
• Physics: calibrate simulation dynamics against real-world observations. ion space of a “full-stack simulation engineer” and jointly considers geometry, materials, colliders, articulation, and physics. It automatically calibrates physics against real trajectories and matches task-specific dynamics.
Simulation environments is useful only if it can provide meaningful policy level signals. Methods, comparisons, and results below.
31
66
580
122,354
Mercedes Bent retweeted
dinner w the design goats! @iansilber, @henrymodis, @paulstamatiou
thank you for organizing a wonderful evening, @mercebent, @veelarco it was truly surreal. 🩵
thank you for organizing a wonderful evening, @mercebent, @veelarco it was truly surreal. 🩵
0
1
4
365
Mercedes Bent retweeted
no one is doing events like surrealbuilders.com
love this community @veelarco and @mercebent are building 🫶
love this community @veelarco and @mercebent are building 🫶
1
1
5
928
Top things I have used Fo for:
1. Real estate agent - messages me 7am and 6pm every day about any new rentals or homes for sale
2. canceled a United flight for me when I was running late in LA + got a credit! it called United
3. Weekend activity planner - kids stuff - so I know whats happening locally
1. Real estate agent - messages me 7am and 6pm every day about any new rentals or homes for sale
2. canceled a United flight for me when I was running late in LA + got a credit! it called United
3. Weekend activity planner - kids stuff - so I know whats happening locally
I led engineering at Google DeepMind.
Today, I'm proud to introduce Fo to give personal AI something no lab ever has... Humans.
Other personal AI's pretend AI can do everything. Fo employs humans to do tasks that AI cannot.
- 2x better at real-world task completion (beats 94% trust rate (4x less likely to leak private info vs Muse, Instinct)
Sign up for free: wajo.ai/join-wajo
4
0
15
5,248
Mercedes Bent retweeted
I led engineering at Google DeepMind.
Today, I'm proud to introduce Fo to give personal AI something no lab ever has... Humans.
Other personal AI's pretend AI can do everything. Fo employs humans to do tasks that AI cannot.
- 2x better at real-world task completion (beats 94% trust rate (4x less likely to leak private info vs Muse, Instinct)
Sign up for free: wajo.ai/join-wajo
Today, I'm proud to introduce Fo to give personal AI something no lab ever has... Humans.
Other personal AI's pretend AI can do everything. Fo employs humans to do tasks that AI cannot.
- 2x better at real-world task completion (beats 94% trust rate (4x less likely to leak private info vs Muse, Instinct)
Sign up for free: wajo.ai/join-wajo
756
678
6,542
2,989,954
Mercedes Bent retweeted
Stanford Prof. @chrmanning on why Moonlake is betting against one giant neural net for its world model, and how coding models changed what’s possible in robot simulation:
"Moonlake is building a world model and simulation platform for robotics deployment and testing."
"There's elopments in coding models, with Claude Code, Codex, and other products. That means you can now be using coding-based AI products to build simulation environments."
"The default has been a room full of robots folding the T-shirt, recording all the interactions, which is very slow, very expensive, and not very scalable. That's a large part of what's held robotics back."
"Some other companies are based around, the whole thing is gonna be this one big neural net. Our bet is that that technology is not ready in terms of physical accuracy, long-term consistency."
"Code-based physics-grounded simulation, overlaid with diffusion models for more accurate material surfaces, can provide the accuracy and the long-term consistency."
@moonlake @Stanford
"Moonlake is building a world model and simulation platform for robotics deployment and testing."
"There's elopments in coding models, with Claude Code, Codex, and other products. That means you can now be using coding-based AI products to build simulation environments."
"The default has been a room full of robots folding the T-shirt, recording all the interactions, which is very slow, very expensive, and not very scalable. That's a large part of what's held robotics back."
"Some other companies are based around, the whole thing is gonna be this one big neural net. Our bet is that that technology is not ready in terms of physical accuracy, long-term consistency."
"Code-based physics-grounded simulation, overlaid with diffusion models for more accurate material surfaces, can provide the accuracy and the long-term consistency."
@moonlake @Stanford
9
11
45
16,674
Mercedes Bent retweeted
Exa is now one of the largest indexes in the world.
We serve 80B pages, track 1.4T urls, and are on track to be Google-scale in early 2027.
It's difficult to know the size of indexes, but we estimate Google to be ~1T, Bing to be ~500B, and Yandex ~200B. Brave mentioned in 0B.
Most of the web is trash that can hurt AI outputs, so you actually have to crawl way more than you serve and train models to filter out the trash.
In terms of peak QPS, we estimate Bing at roughly 30k/s and Google at ~500k/s.
In our case, because agent traffic often requires large fan-out (deep searches can use dozens of sub-searches) and can be spiky (like when AI labs RL with us), we're already starting to provision for Bing-scale traffic.
Within 2 years, agents will search at many millions per second. They'll also want comprehensiveness over all data, far bigger than any index today. So the retrieval infra required will be larger than Google-scale in both dimensions.
We serve 80B pages, track 1.4T urls, and are on track to be Google-scale in early 2027.
It's difficult to know the size of indexes, but we estimate Google to be ~1T, Bing to be ~500B, and Yandex ~200B. Brave mentioned in 0B.
Most of the web is trash that can hurt AI outputs, so you actually have to crawl way more than you serve and train models to filter out the trash.
In terms of peak QPS, we estimate Bing at roughly 30k/s and Google at ~500k/s.
In our case, because agent traffic often requires large fan-out (deep searches can use dozens of sub-searches) and can be spiky (like when AI labs RL with us), we're already starting to provision for Bing-scale traffic.
Within 2 years, agents will search at many millions per second. They'll also want comprehensiveness over all data, far bigger than any index today. So the retrieval infra required will be larger than Google-scale in both dimensions.
169
230
3,622
943,477
Mercedes Bent retweeted
At @cadre, I saw the gap firsthand. Large “institutional” firms built their own systems. But everyone else had to deliver with fewer resources, spreadsheets and manual work.
forbes.com/sites/jeffkauf…
“Institutional” should describe a standard, not a budget.
That conviction became @ellis_company
@JeffKauflin, @Forbes Ryan Williams’ Next Act After Cadre: A Fintech Fix For Private Credit’s Data Mess
forbes.com/sites/jeffkauf…
“Institutional” should describe a standard, not a budget.
That conviction became @ellis_company
@JeffKauflin, @Forbes Ryan Williams’ Next Act After Cadre: A Fintech Fix For Private Credit’s Data Mess
13
11
87
1,344,695
Mercedes Bent retweeted
What comes after language?
Join us at AGI Summit 2026 for “Beyond Language”—a conversation on world models, spatial reasoning, and embodied intelligence.
Featuring Christopher Manning, Mercedes Bent, Fan-Yun Sun, and Christian Laforte.
@chrmanning @mercebent @sunfanyun @chrlaf ace of Fine Arts, San Francisco
🎟️ agisummit.ai
#AGISummit #ArtificialIntelligence #AGI #WorldModels #EmbodiedAI #SpatialReasoning
Join us at AGI Summit 2026 for “Beyond Language”—a conversation on world models, spatial reasoning, and embodied intelligence.
Featuring Christopher Manning, Mercedes Bent, Fan-Yun Sun, and Christian Laforte.
@chrmanning @mercebent @sunfanyun @chrlaf ace of Fine Arts, San Francisco
🎟️ agisummit.ai
#AGISummit #ArtificialIntelligence #AGI #WorldModels #EmbodiedAI #SpatialReasoning
4
4
15
2,186
Mercedes Bent retweeted
Why are you paying for dictation?
We're releasing free, unlimited AI dictation.
And it's not a slow, local model.
Willow Frontier Mini is cloud-based with zero-data-retention. More accurate and faster than Wispr Flow, OpenAI, Deepgram, and more.
See video comparison.
We're releasing free, unlimited AI dictation.
And it's not a slow, local model.
Willow Frontier Mini is cloud-based with zero-data-retention. More accurate and faster than Wispr Flow, OpenAI, Deepgram, and more.
See video comparison.
251
114
2,145
919,446
Mercedes Bent retweeted
Introducing Willow Frontier Mini and Willow Frontier Pro.
Frontier Mini gives everyone free, unlimited dictation.
Frontier Pro is our most accurate model yet, built for power users and teams who want the fastest and best dictation experience.
Frontier Mini gives everyone free, unlimited dictation.
Frontier Pro is our most accurate model yet, built for power users and teams who want the fastest and best dictation experience.
175
62
940
488,094
Mercedes Bent retweeted
We raised $4.3M, led by @PanteraCapital
Soon, there will be more agents than people online. Those agents will book flights, hire contractors, enrich customer data, conduct research, and complete transactions without a human in the loop. But today, agents are stuck with whatever lly given. The moment they need something new, they fail, hallucinate, or hand the problem back to a human.
Orthogonal fixes this. Through a single integration, agents can discover the capabilities they need in the moment, orchestrate them, and pay for them instantly. An agent describes what it wants, and Orthogonal composes the result, calling the right services in the right order.
Our goal is simple: when an agent needs a capability it doesn't already have, Orthogonal will be the first place it goes.
Thanks to @PanteraCapital , @ycombinator, @pioneer_fund, @decasonic, Blast Club, Outbound Capital, Rice Capital (@taro_f), Surreal by Premise (@mercebent & @veelarco), Batch Ventures (CTO Fund), and our strategic investors for backing us. We're building the default front door for the internet.
Soon, there will be more agents than people online. Those agents will book flights, hire contractors, enrich customer data, conduct research, and complete transactions without a human in the loop. But today, agents are stuck with whatever lly given. The moment they need something new, they fail, hallucinate, or hand the problem back to a human.
Orthogonal fixes this. Through a single integration, agents can discover the capabilities they need in the moment, orchestrate them, and pay for them instantly. An agent describes what it wants, and Orthogonal composes the result, calling the right services in the right order.
Our goal is simple: when an agent needs a capability it doesn't already have, Orthogonal will be the first place it goes.
Thanks to @PanteraCapital , @ycombinator, @pioneer_fund, @decasonic, Blast Club, Outbound Capital, Rice Capital (@taro_f), Surreal by Premise (@mercebent & @veelarco), Batch Ventures (CTO Fund), and our strategic investors for backing us. We're building the default front door for the internet.
78
24
363
77,294