5.4 is really special.
Daily Edition
2026-03-11
X / Twitter
16There are some pretty wild downstream effects in a world with trillions of agents using the internet and software.
One very big one is what happens with agents with budgets and wallets. There are lots of business models that never ended up working out for the human-based internet that all of a sudden start to make economic sense in an agent-based internet.
Think of all the proprietary data and research that’s sitting out there right now behind a paywall that a human will never run into. Finance data, medical research, and so on. Most people won’t sign up for a $100 or $1000 subscription for information they need infrequently. The cost is too high. Equally, micropayments for this data rarely worked at scale because the volume was too low to matter.
However, now an agent can have a budget for a specific set of research it’s doing, and the agent might pay $0.1 or $1 to access it in a workflow. And now that data may be relevant in 1,000X’s more use-cases than it was before.
Similarly, there are many APIs and tools out there on the web that don’t make sense to have a subscription for, but now an agent may interact with for a specific exchange, and it could cost $.01 or $0.1 per transaction.
All of a sudden new kinds of software can get built and monetized that would have been uneconomical before. Some new form of commercial open source, essentially.
Obviously lots of infrastructure and agreement across the industry is needed for this -and getting discovered by the agent is going to be a whole new class of search and discovery problem- but there are so many potentially interesting new scenarios here.
François Chollet: AI agents will soon graduate to fully-fledged economic actors that buy services, compute, and even data in the course of accomplishing high-level goals. 1-2 years before we start seeing this at scale.
One very big one is what happens with agents with budgets and wallets. There are lots of business models that never ended up working out for the human-based internet that all of a sudden start to make economic sense in an agent-based internet.
Think of all the proprietary data and research that’s sitting out there right now behind a paywall that a human will never run into. Finance data, medical research, and so on. Most people won’t sign up for a $100 or $1000 subscription for information they need infrequently. The cost is too high. Equally, micropayments for this data rarely worked at scale because the volume was too low to matter.
However, now an agent can have a budget for a specific set of research it’s doing, and the agent might pay $0.1 or $1 to access it in a workflow. And now that data may be relevant in 1,000X’s more use-cases than it was before.
Similarly, there are many APIs and tools out there on the web that don’t make sense to have a subscription for, but now an agent may interact with for a specific exchange, and it could cost $.01 or $0.1 per transaction.
All of a sudden new kinds of software can get built and monetized that would have been uneconomical before. Some new form of commercial open source, essentially.
Obviously lots of infrastructure and agreement across the industry is needed for this -and getting discovered by the agent is going to be a whole new class of search and discovery problem- but there are so many potentially interesting new scenarios here.
François Chollet: AI agents will soon graduate to fully-fledged economic actors that buy services, compute, and even data in the course of accomplishing high-level goals. 1-2 years before we start seeing this at scale.
When @thisisgrantlee was pitching early investors for @GammaApp, one called it "the worst idea I've ever heard" and then hung up.
Grant and Gamma (valued $2.1B) are doing ok. No word on the investor.
Full talk out tomorrow @southpkcommons.
Grant and Gamma (valued $2.1B) are doing ok. No word on the investor.
Full talk out tomorrow @southpkcommons.
Companies like Meta subsidize third party developers in various ways to help grow their platforms, then take 30% of the developer revenue right back with the platform tax, which is a wasteful churn.
Epic’s tax structure of zero for the first $1M / year in revenue seems like a clean solution. You would still need explicit subsidies to get certain types of games / apps created at all, but it perfectly rewards what you actually want: increased economic activity, versus a biased pre-selection process.
@TimSweeneyEpic Have there been any unexpected perverse incentives as a result of this policy?
If it wasn’t so easily exploited by buying your own app, a negative rate “earned income tax credit” for initial revenue would actually be a good incentive for a platform like Quest.
Epic’s tax structure of zero for the first $1M / year in revenue seems like a clean solution. You would still need explicit subsidies to get certain types of games / apps created at all, but it perfectly rewards what you actually want: increased economic activity, versus a biased pre-selection process.
@TimSweeneyEpic Have there been any unexpected perverse incentives as a result of this policy?
If it wasn’t so easily exploited by buying your own app, a negative rate “earned income tax credit” for initial revenue would actually be a good incentive for a platform like Quest.
Companies like Meta subsidize third party developers in various ways to help grow their platforms, then take 30% of the developer revenue right back with the platform tax, which is a wasteful churn.
Epic’s tax structure of zero for the first $1M / year in revenue seems like a clean solution. You would still need explicit subsidies to get certain types of games / apps created at all, but it perfectly rewards what you actually want: increased economic activity, versus a biased pre-selection process.
@TimSweeneyEpic Have there been any unexpected perverse incentives as a result of this policy?
If it wasn’t so easily exploited by buying your own app, a negative rate “earned income tax credit” for initial revenue would actually be a good incentive for a platform like Quest.
Epic’s tax structure of zero for the first $1M / year in revenue seems like a clean solution. You would still need explicit subsidies to get certain types of games / apps created at all, but it perfectly rewards what you actually want: increased economic activity, versus a biased pre-selection process.
@TimSweeneyEpic Have there been any unexpected perverse incentives as a result of this policy?
If it wasn’t so easily exploited by buying your own app, a negative rate “earned income tax credit” for initial revenue would actually be a good incentive for a platform like Quest.
Amjad is a legendary programmer and founder.
It's been amazing to see him build out this amazing product and company.
Amjad Masad: Software isn’t merely technical work anymore. It’s creative.
Introducing Replit Agent 4. The first AI built for creative collaboration between humans and agents.
Design on an infinite canvas, work with your team, run parallel agents, and ship working apps, sites, slides & more.
It's been amazing to see him build out this amazing product and company.
Amjad Masad: Software isn’t merely technical work anymore. It’s creative.
Introducing Replit Agent 4. The first AI built for creative collaboration between humans and agents.
Design on an infinite canvas, work with your team, run parallel agents, and ship working apps, sites, slides & more.
Our imaginations are unbounded, so should the worlds we create be…🌐😍
World Labs: A single image can transform into a massive world composed of millions of splats with Marble. Forming large explorable environments from ancient ruins to spaceship interiors.
World Labs: A single image can transform into a massive world composed of millions of splats with Marble. Forming large explorable environments from ancient ruins to spaceship interiors.
Our imaginations are unbounded, so should the worlds we create be…🌐😍
World Labs: A single image can transform into a massive world composed of millions of splats with Marble. Forming large explorable environments from ancient ruins to spaceship interiors.
World Labs: A single image can transform into a massive world composed of millions of splats with Marble. Forming large explorable environments from ancient ruins to spaceship interiors.
Excited to release PostTrainBench v1.0!
This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting.
We believe this is a first step toward tracking progress in recursive self-improvement 🧵:
This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting.
We believe this is a first step toward tracking progress in recursive self-improvement 🧵:
Excited to release PostTrainBench v1.0!
This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting.
We believe this is a first step toward tracking progress in recursive self-improvement 🧵:
This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting.
We believe this is a first step toward tracking progress in recursive self-improvement 🧵:
My autoresearch labs got wiped out in the oauth outage. Have to think through failovers. Intelligence brownouts will be interesting - the planet losing IQ points when frontier AI stutters.
http://x.com/i/article/2031773086690013184
Claude for all things knowledge work feels like it is on a very similar trajectory to what agentic coding experienced last year.
I expect entire industries that rely on spreadsheets and powerpoints to begin to be transformed in the next few months.
Claude: Claude for Excel and Claude for PowerPoint now sync together seamlessly.
When you’ve got more than one file open, Claude shares the full context of your conversation between them.
Pull data from spreadsheets, build out tables, and update a deck — without re-explaining a step.
I expect entire industries that rely on spreadsheets and powerpoints to begin to be transformed in the next few months.
Claude: Claude for Excel and Claude for PowerPoint now sync together seamlessly.
When you’ve got more than one file open, Claude shares the full context of your conversation between them.
Pull data from spreadsheets, build out tables, and update a deck — without re-explaining a step.
Claude for all things knowledge work feels like it is on a very similar trajectory to what agentic coding experienced last year.
I expect entire industries that rely on spreadsheets and powerpoints to begin to be transformed in the next few months.
Claude: Claude for Excel and Claude for PowerPoint now sync together seamlessly.
When you’ve got more than one file open, Claude shares the full context of your conversation between them.
Pull data from spreadsheets, build out tables, and update a deck — without re-explaining a step.
I expect entire industries that rely on spreadsheets and powerpoints to begin to be transformed in the next few months.
Claude: Claude for Excel and Claude for PowerPoint now sync together seamlessly.
When you’ve got more than one file open, Claude shares the full context of your conversation between them.
Pull data from spreadsheets, build out tables, and update a deck — without re-explaining a step.
1️⃣ 𝚗𝚙𝚖 𝚒 𝚌𝚑𝚊𝚝
2️⃣ 𝚌𝚛𝚎𝚊𝚝𝚎𝚆𝚑𝚊𝚝𝚜𝙰𝚙𝚙𝙰𝚍𝚊𝚙𝚝𝚎𝚛()
3️⃣ There's no step 3
Easiest way to start building WhatsApp agents
Vercel Developers: Chat SDK now supports WhatsApp.
Extend your agentic chat bots with WhatsApp messages, reactions, typing indicators, location sharing, and more using a simple adapter:
𝚊𝚍𝚊𝚙𝚝𝚎𝚛𝚜: {
𝚠𝚑𝚊𝚝𝚜𝚊𝚙𝚙: 𝚌𝚛𝚎𝚊𝚝𝚎𝚆𝚑𝚊𝚝𝚜𝙰𝚙𝚙𝙰𝚍𝚊𝚙𝚝𝚎𝚛()
}
https://vercel.com/changelog/chat-sdk-adds-whatsapp-adapter-support
2️⃣ 𝚌𝚛𝚎𝚊𝚝𝚎𝚆𝚑𝚊𝚝𝚜𝙰𝚙𝚙𝙰𝚍𝚊𝚙𝚝𝚎𝚛()
3️⃣ There's no step 3
Easiest way to start building WhatsApp agents
Vercel Developers: Chat SDK now supports WhatsApp.
Extend your agentic chat bots with WhatsApp messages, reactions, typing indicators, location sharing, and more using a simple adapter:
𝚊𝚍𝚊𝚙𝚝𝚎𝚛𝚜: {
𝚠𝚑𝚊𝚝𝚜𝚊𝚙𝚙: 𝚌𝚛𝚎𝚊𝚝𝚎𝚆𝚑𝚊𝚝𝚜𝙰𝚙𝚙𝙰𝚍𝚊𝚙𝚝𝚎𝚛()
}
https://vercel.com/changelog/chat-sdk-adds-whatsapp-adapter-support
Software will look very different in the future for most knowledge work. Our software was designed for people to do most of the work, but now agents will be doing a large portion of those tasks and don’t need any of the same UI.
This means our software primarily will be about managing the agents doing that, intervening when they go in the wrong direction, giving them the right context, integrating that work into a broader workflow, being able to edit and manipulate the final output, and so on.
Lots of change to come as agents get more and more powerful.
Andrej Karpathy: Expectation: the age of the IDE is over
Reality: we’re going to need a bigger IDE
(imo).
It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming.
This means our software primarily will be about managing the agents doing that, intervening when they go in the wrong direction, giving them the right context, integrating that work into a broader workflow, being able to edit and manipulate the final output, and so on.
Lots of change to come as agents get more and more powerful.
Andrej Karpathy: Expectation: the age of the IDE is over
Reality: we’re going to need a bigger IDE
(imo).
It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming.
YouTube
1
Unsupervised Learning: Redpoint's AI Podcast