Fintech CRO at F2 (YC)💰📈Unicorn builder 🦄 🔨 Winemaker @ Coteau Pavot 🍇 proud dad
Joined March 2015
- Tweets 3,460
- Following 3,269
- Followers 1,658
- Likes 41,116
174 Photos and videos
Today, the meter starts for everyone else. As of June 15, Anthropic moved off flat-rate subscriptions.
Automation now runs on a fixed monthly credit. And once it's gone, every token meters at full API rates.
"Tokenmaxxing" was a strategy that only made sense when tokens felt free.
They were never free. They were subsidized. The subsidy just ended.
Here's the part every CFO is about to internalize: the capability gap between models is closing. The pricing gap is not.
So every enterprise now faces the same fork in the road.
Stay in the metered gridlock of single-lab lock-in: where one vendor sets your price, your rate limits, and your access, and changes all three whenever it suits them.
Or take the off-ramp: route every task to the most cost-efficient model that does it well, keep your context and memory in your own hands, and never depend on one lab again.
The companies that take the off-ramp will use more AI than ever, for less, and on their own terms.
Freedom from the frontier.
1
2
3
427
“the real opportunity is not in picking the best model but instead in building a learning loop on top of models where human capital and token capital compound”
couldn’t agree more @satyanadella 🐐
1
3
256
SITUATION DETECTED: Microsoft is limiting internal employee use of Fable 5 over Anthropic's new data retention requirements, per The Verge.
Fable 5 requires data retention to operate its safety classifiers, unlike other Claude models which run under Zero Data Retention rules.
9
29
640
68,644
This is spot on. The open-vs-closed capability gap is collapsing faster than the pricing gap.
Inside the enterprise, most teams still reach for the most expensive frontier model by default with “zero governance, auditability and control.”
I ran a like-for-like comparison of flagship model costs below. That token-maxxing behavior is starting to change quickly in enterprise, as we’ve already seen with Uber and Microsoft.
My prediction? As intelligence commoditizes, open-weight models keep improving, and vertical harnesses outperform on agentic workflows, the value will accrue to the LLM-agnostic multi-model application layer.
@chamath would love to get your take for my Forbes column on this topic dropping Monday.
Jun 6
Your margin is my opportunity: AI version…
The biggest surprise of 2026 is that the capability gap between the best open-weight/source models and the best closed models has narrowed much faster than the pricing gap. The pricing gap remains enormous while the capability gap is quite narrow.
What does this means in practice?
For a company consuming 1 billion input tokens and 1 billion output tokens per month:
GPT-5.5 Pro: ~$105,000
Claude Opus 4.8: ~$30,000
DeepSeek V4 Pro: ~$5,220
DeepSeek R1: ~$2,740
I asked ChatGPT what it thought about this and it answered as follows:
“If I were building a company today, the economic frontier would look roughly like:
DeepSeek V4 Pro / R1 for high-volume inference.
Claude Opus for premium agent workflows where reliability matters.
GPT-5.5 Pro only for workloads where its incremental capability demonstrably produces enough business value to justify a 20–40× token premium.”
Most CEOs have no idea that, instead of this nuanced approach, their teams are running amok internally by picking the most expensive models in most cases and burning through massive budgets with zero governance, audit ability and control.
As control planes like our Software Factory become more standard, you can expect the run rate revenue growth of the frontier labs to go down meaningfully and the revenues of the open models to skyrocket.
Why? Because we can implement the nuanced approach above and be agnostic to model - instead focusing on customer intent, model task and cost management among other things.
1
1
6
1,045
10 months ago we spun F2 out of Arc with a $10M pre-seed. Nascent tech, no traction. Today: live across 100 credit funds and banks (Bain Capital, Live Oak). MAUs up 650%. Deals up 10x YTD. We just raised $14M to double down.
Come Build with us.
6
6
67
752,160
Basile Senesi retweeted
Jun 10
Seems clear to me that being model agnostic is the best move for finance specific AI companies.
Btw, f2 is a really clever name here.
10 months ago we spun F2 out of Arc with a $10M pre-seed. Nascent tech, no traction. Today: live across 100 credit funds and banks (Bain Capital, Live Oak). MAUs up 650%. Deals up 10x YTD. We just raised $14M to double down.
Come Build with us.
3
3
45
27,784
Basile Senesi retweeted
apparently there's no zero data retention for Claude Fable 5
10
3
163
309,982
Basile Senesi retweeted
Jun 9
yup. non-starter for enterprise customers
apparently there's no zero data retention for Claude Fable 5
23
32
2,292
254,328
.@DanielleFong explains how drones could be used for fighting fires:
"Think of the LA fire... people love to build right up next to hilly brush land that's extremely flammable. You can't reach it with trucks."
"With a continuous launching of drones, you could put out fires in any of those areas."
"When you increase the range of firefighting drones by a factor of 20 or 30, you increase the amount of area you can protect with the same drones by 400x to 900x."
"Convective Capital estimates the total health burden just from wildfires on the West Coast at about $100 billion a year."
5
4
48
8,103
Basile Senesi retweeted
May 29
my heuristics in life is buying whatever the top 1% most frequent user buy, because their lives rely on the good working most of the time
most NYC/CHI pilots and flight attendants use Travelpro (and they use transit to the airport) soft top rollaboard. Platinum Elite is great
It's not just luxury hotels that are a scam, it's almost everything that's luxury that's a scam
Gf bought Rimowa suitcases, expensive and supposed to be better quality than regular ones, but of course they're much worse
They keep breaking, like all of them, cracks in the handles, cracks in the sides, it's just cheap plastic shit but it costs $1000 or more
Rimowa was bought by LVMH in 2016 which has an average profit margin of 66% and whose strategy is to increase prices by ~5x, decrease costs by ~5x and then create artificial scarcity (limited availability per shop) because people want what they can't get (not me though but many)
LVMH is kinda like the luxury version of private equity, it makes everything more expensive and worse and hard to get!
90
209
7,216
1,297,223
Basile Senesi retweeted
May 28
I think they are going to be looking for more AI enabled GTM folks soon. The more I automate, the more I want to do, not less
1
1
5
1,840
Basile Senesi retweeted
May 27
Get that bag brother. Hebbia is ass tho
2
8
188
Basile Senesi retweeted
May 21
Most large companies are multi model by default. They use both openai for some workflows, anthropic for others, including oss and they benchmark constantly. AI is not winner take all saas yet
1
1
102
21,875
Basile Senesi retweeted
AI-driven knowledge management is starting to take shape as a real category in private markets.
Excited to see how @F2aiHQ's launch of Institutional Knowledge could help firms operationalize the pattern recognition historically locked in memos, models, and team experience.
Private markets firms have decades of pattern recognition trapped in memos, models, and churned associates heads.
Today, F2 launched Institutional Knowledge to solve this problem for our ~100 private market clients.
IK turns that history into a persistent intelligence layer, so every deal makes the next one sharper.
DM for a demo.
@F2aiHQ
1
4
340
Basile Senesi retweeted
May 6
Firing all of my analysts (except for @wallstreetkarl) and having him crank out investment memos and models with AI
Private markets firms have decades of pattern recognition trapped in memos, models, and churned associates heads.
Today, F2 launched Institutional Knowledge to solve this problem for our ~100 private market clients.
IK turns that history into a persistent intelligence layer, so every deal makes the next one sharper.
DM for a demo.
@F2aiHQ
4
1
44
32,365
Private markets firms have decades of pattern recognition trapped in memos, models, and churned associates heads.
Today, F2 launched Institutional Knowledge to solve this problem for our ~100 private market clients.
IK turns that history into a persistent intelligence layer, so every deal makes the next one sharper.
DM for a demo.
@F2aiHQ
8
8
78
709,183
May 5
Most private markets firms have years of judgment trapped in past IC memos, models, diligence notes, and deal files.
@F2aiHQ Institutional Knowledge turns that history into a queryable intelligence layer, so teams can underwrite faster and with more context.
Every deal makes the next one sharper.
2
85
Basile Senesi retweeted
Apr 30
Today's dads do 2x the childcare their fathers did. Where do they find the time? Mainly: they work less. But also: they spend less time relaxing (TV, reading, and other leisure). [1/5]
27
233
2,278
375,392