Pierre de la Grand'rive

377 posts

Pierre de la Grand'rive banner
Pierre de la Grand'rive

Pierre de la Grand'rive

@pierre_dlgr

CEO @ Delos | Building AGI

Paris Katılım Mart 2024
136 Takip Edilen2K Takipçiler
Sabitlenmiş Tweet
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
Introducing Workers: unlimited AI employees that run your company... And we've just made $1M ARR in a couple of days. Most knowledge jobs shouldn't exist. Not because the work isn't valuable. Because the way we execute it is absurd. Millions of people spend their days: - Answering emails - Following up - Updating CRM - Coordinating meetings - Moving information between systems There is no expertise here. That's workflow.. Workers are AI employees that can perform work across your organization like a human teammate. The line between software and labor just disappeared.
English
224
180
1K
1.2M
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
AI will 10x global GDP. Almost nobody reading this will get 10x richer. The multiplier goes to whoever puts it to work, not whoever owns the model.
English
0
0
5
185
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
the burn is the point. nobody built anything great by asking for ROI first
Vaibhav Sisinty@VaibhavSisinty

I just came across a Substack article that shifts the entire AI spending conversation. The number it puts on the board is $1.5 trillion. 🤯 In 2023, Sequoia asked the question: where is the $200 billion in AI revenue to justify the spend? In 2024 it became $600 billion. In 2025 it was $840 billion. In July 2026 it is now $1.5 trillion. The gap is not closing. It is accelerating. Here is the math the article talks about. → Big tech companies like Microsoft, Google, Amazon, and Meta are spending roughly $750 billion this year building data centers and buying chips for AI. → But those companies do not sell AI directly to you. They sell it to OpenAI, Anthropic, and others, who then sell it to customers. Two layers. Each layer needs to make money. → So the end customer revenue needed to make this all work is roughly double. That is $1.5 trillion. In one year. → And this adds up. Every year since ChatGPT launched, the industry has been building more. → Add it all together and the total revenue needed to justify everything built so far is roughly $3 trillion. Nobody has come close to generating that. Revenue is growing. Anthropic went from $0 to $60 billion in ARR. That is genuinely historic. But the infrastructure spend is growing faster. Memory is already sold out for the next two years. Micron just broke ground on a $9.3 billion HBM expansion in Hiroshima. Apple raised MacBook prices by $300 because of chip costs. Consumers are last in line. And here is what nobody is asking. The math requires $1.5 trillion in revenue this year alone. The two most successful AI companies on the planet combined generate less than $85 billion. Who is paying the remaining $1.4 trillion? The question is no longer whether AI works. It clearly does. The question is whether the economics work before the runway empties.

English
0
1
2
503
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
we swapped our stack to GLM 5.2 and watched the inference bill drop 86% overnight. same output, one line in the config. the frontier models aren't 7x better anymore, they're just 7x more expensive.
Pierre de la Grand'rive tweet media
English
0
0
1
151
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
Multi party autonomous agents are the actual frontier. Not one agent that writes code or answers tickets. Agents that negotiate with each other, across company boundaries, with their own identity and decision rights. Lyzr just let an AI agent run a $100M fundraising round. Fielded 130+ investor questions, drafted memos, tracked which slides people lingered on. $400M in interest. The founder barely showed up. That's what's coming. But everyone is obsessed with the wrong infrastructure. Google's A2A protocol wants agents to discover each other through standardized APIs and protocol cards. It's too early. Autonomous agents won't collaborate through protocols. They'll collaborate like humans do: via email, Slack, visio, shared docs. You don't need a protocol when you have a mailbox. The real bottleneck isn't connectivity. It's memory. Multi agent systems fail 41 to 86% of the time in production. The top failure mode isn't the model. It's context loss during handoffs. Agents forget what was decided two conversations ago. "Just store the conversation" doesn't work. Context degrades at scale. Memory is what makes multi party agents actually viable. You tell an agent something once, and it remembers when it matters. Not a database. Not infinite context. Just: it was in the meeting, it understood what was decided, and it brings that to the next one. That's what human teams do. Agents need the same. Everyone is building agents. Almost nobody is building the coordination layer. The companies that figure out how agents actually work together will eat the ones that don't. Not because their agents are smarter. Because their agents can actually cooperate.
English
0
1
3
231
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
the most powerful AI company on earth is running out of imagination. Fidji Simo just stepped down. she was Altman's #2, CEO of AGI deployment. the person whose entire job was figuring out what to actually DO with AGI. now she's a part time advisor. the person who was supposed to imagine the future of AI at the company building AGI is gone. they killed Atlas. their browser. the one product that imagined a new way to use the internet when an agent is the starting point. folded into a Chrome extension. the most imaginative product in their pipeline, killed. GPT 5.6 dropped yesterday. three tiers: Sol, Terra, Luna. the big innovation is pricing tiers. multi agent beta that reads like a checkbox on a feature list. nobody at OpenAI seems excited about what they're shipping. they're shipping because the calendar says it's time, not because someone dreamed up something new. Sora is getting trimmed too. Fidji's memo called them side quests. but the side quests were the imagination. the browser, the video model, the weird experiments. that's where the next thing was going to come from. they're cutting the exact things where imagination lives. they filed confidentially for IPO in June. $1T target valuation. the entire org is now oriented around looking good on a roadshow. every product decision filtered through will this look clean in the IPO filing? you don't kill your browser and launch pricing tiers because you're bold. you do it because imagination is a liability on a roadshow. OpenAI has more compute, more talent, more money, more users than anyone. more data, more brand, more distribution. every possible advantage. what they don't have is someone in the room who's still dreaming. the most valuable AI company in the world is playing it safe and calling it discipline.
Sam Altman@sama

i am really sad about this and very grateful for all fidji has done for openai, and even grateful for her friendship and who she is as a person. we all wish her the best for a speedy recovery. this sucks.

English
0
0
2
509
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
@forgebitz Coding was never the bottleneck. Specifying what to build was. AI writes code faster but still can't figure out what you actually want.
English
0
0
2
76
Klaas
Klaas@forgebitz·
i am still somewhat skeptical about ai coding the models are really really good now, and every company has access but software seems more sloppy and broken than ever before as if coding was never really the bottleneck
English
117
6
406
33K
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
@davep Two pizzas was never about the pizzas. It was about coordination cost. AI doesn't remove coordination, it just moves it upstream.
English
0
0
0
159
David Pan
David Pan@davep·
RIP to the two pizza team. Jeff Bezos gave us an all-time great metaphor: keep the team small enough to feed with two pizzas. It shaped how engineering orgs got built for 20+ years. He was right about small teams. Still is. But in the AI era, two pizzas is too much pizza.
English
173
67
2.6K
310.2K
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
@tszzl the framing matters less than the deployment gap. most companies still cant get tool AI to work
English
0
0
0
101
Ethan Mollick
Ethan Mollick@emollick·
I can basically guarantee you are not being ambitious enough with the work you are assigning Fable Start asking for the maximum possible thing to figure out the far edges of what it can do. After that, you can decide where the system reaches it limits & revise requests downward
English
142
79
2K
131.5K
Pierre de la Grand'rive
Pierre de la Grand'rive@pierre_dlgr·
@_K_Stiles 6,000 engineers just to make Copilot work 😅 and that's still nowhere near enough😅 and that's still nowhere near enough
English
0
0
1
6
Katherine Stiles
Katherine Stiles@_K_Stiles·
Microsoft, AWS deploy engineer armies to help crack AI AWS and Microsoft want to embed thousands of their own engineers at client companies to help them capitalize on artificial intelligence, which has yet to prove its profitability in the business world. Microsoft announced Thursday the creation of a unit called Microsoft Frontier Company, which is backed by a $2.5 billion investment and brings together 6,000 experts and engineers. More: techxplore.com/news/2026-07-m… @TechXplore_com
Katherine Stiles tweet media
English
1
0
4
67