The Week the AI Story Shifted

Today on the AI Daily Brief, we're discussing a week in which the AI story shifted, or at least started to fork. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Granola, Robots and Pencils, and ZenCoder. To get an ad-free version of the show, go to patreon.com/aiDailyBrief, or you can subscribe on Apple Podcasts. If you want to learn more about sponsoring the show, send us a note at sponsors@aidailybrief.ai. Aidailybrief.ai is also where you are going to find out about everything going on in the community. Right now, I am continuing to point people to the April AI Usage Pulse survey. And if you want to see why it's valuable to share this information, go check out pulse.aidailybrief.ai. And of course, there'll be a link to that in the show notes as well. In it, you can see the individual monthly responses from the last 3 months of AI Usage Pulse surveys, as well as the big overarching trends, like the growth in agentic use cases. The April survey is now available. You can do it right there. And if you complete this, you will get the results before everyone else. Now last week I told you about an experiment that I was going to be trying where if Friday happened to be a comparatively slow day in AI, nothing is actually slow, I was going to start experimenting with some sort of weekly recap. The goal of the weekly recap is not just to rehash the same stories we talked about, but to put them in an overarching context that helps you understand in just 20 or 25 minutes what the big point of that week was. For people who aren't able to listen as much, it's a way to, in a single episode, have the broad brushstrokes of what happened, and for folks who are daily listeners, it's a chance to reinforce the themes that you've been hearing all week. Now, I was very positively pleased with the response. A lot of you provided great feedback, and the numbers also suggest that this is a valuable type of episode to at least consider. I'm not sure that it'll be every week, and I think probably on some weeks I will need to use the open slot on Saturday for this, given that there will often be news that we need to cover in a normal form. But for now, we're gonna do another weekly recap. And if last week was the week that AI grew up, with the thesis of that episode being that we were starting to see a real maturation of the way that people were engaging with AI on a usage basis, in markets, and more. This is almost a part 2 in some ways where that new maturity started to diffuse into the stories that we were telling about AI, as well as the type of product priorities we have in the launches. The week kicked off on Sunday with this opinion post from Ezra Klein about why the AI job apocalypse probably won't happen, at least in the way that the most fearful folks have been suggesting for some time now. The main inspiration for Ezra in that post was the Alex Emas essay, What Will Be Scarce, that we read a few weeks ago on Long Read Sunday. Now, Alex is a Chicago Booth economist, and the point that he's making in this piece is that when one sector of the economy gets disrupted, new surplus usually flows somewhere. It doesn't just dissipate. The big thing that Alex focuses on in his piece is the idea of the relational sector, where the value of a good or a service that we're consuming is not just based on the good or service itself, but in its particular mode of creation or transmission. In other words, where it matters who made the thing or who's providing the service in what way. Alex argues that the relational sector definitionally will not be affected by AI in the same way as other parts of the economy and will indeed be the recipient and will see a proportional increase in its demand. Now, as an aside, if you're interested in blowing out this argument even more, you are gonna wanna come back for LRS. It is my most full-throated exploration ever of not only why I don't think there's going to be a job apocalypse, but the specific shape and texture of the type of jobs that I think are going to be created. In the wake of AI. So come look back for that. But the point coming back to Ezra Klein is that Alex Yamas's essay was not just floating around Twitter, but actually found its way to mainstream discourse. And frankly, mainstream discourse that was not of the group that has been predisposed to actually being sympathetic to AI or tech in general. Ezra Klein is someone who had AI doomer-in-chief Eliezer Yudkowsky on his show last October. In February, he interviewed Anthropic's Jack Clark in a show that he titled, "How Fast Will AI Agents Rip Through the Economy?" which is what I mean when I say that this essay represents a shift in the story and a shift in the vibe. Now, a much more inside baseball type of source exploring similar themes came from A16z's David George. Why I say, of course, that this one is much more inside is that A16z chief Mark Andreessen has been one of the loudest and most vocal AI accelerationists, and so it's not surprising to see this particular essay titled "The AI Job Apocalypse Is a Complete Fantasy" coming from someone in his organization. Still, what David provides in the piece and why it got resonance beyond just dyed-in-the-wool accelerationists is that in addition to a lot of opinion, there's also a lot of data in here. One of the charts he shows, which if you're listening is very worth going and looking at, is a chart of US employment by sector since 1850. It shows the decrease in agriculture employment from just under 70% back in 1850 down to less than 5% today. And more than that, it shows that since 1950, While a couple of areas, most notably manufacturing and construction, have gone consistently down, the real story is a diversification of the labor market into lots of different sectors. Leisure and hospitality, private education and health, professional and business services. A lot of things that didn't really or barely existed back in the middle of last century. He also shows a bunch of Jevons' Paradox-y type of charts where a thing that you think would have a negative impact actually had a very similar and opposite positive impact. More productive farming, for example, he pointed out, didn't lead to more farmers, but it did lead to more workers as the world was able to take advantage of more productive farming to get a population boom as the world could simply support more people. Another example he points to is the shift in the number of bookkeepers and accounting clerks in the wake of the introduction of the spreadsheet. While those two particular jobs would see a steady decline for the next 30 years, other related areas that were enabled by spreadsheets took off, including financial analysts and accountants and auditors. He also points out that productivity gains don't just make existing services cheaper and more accessible to a different set of people, but also that they lead to entirely new categories categories of services that take advantage of that labor surplus. Nail salons, pet care, exam prep and tutoring, and athletes, coaches, umpires, and related work all had less than 100,000 workers in 1990, and now each have between 150,000 and 350,000. He points to the charts that we've started to see that suggest that the demand for software engineers is rising. And I think in an important chart notes that in mentions of AI workforce impact on public market earnings calls, Augmentation outmentions substitution by a ratio of 8 to 1. Now the real exciting thing to me and the thing that I'll be exploring in the LRS episode I was just mentioning are the jobs that don't exist yet that get enabled by these changes. But you can see why people latched onto David's piece and why it came at a perfect time following Ezra's piece and Alex's piece before that. Now what's interesting is that one of the things that I think is happening is not just blind optimism and all of a sudden people who weren't excited about AI before being excited about it now, but a maturation in our understanding of how the diffusion is actually likely to take place. I think there are a couple parts of that. Another big story from the beginning of the week was that both Anthropic and OpenAI were launching massive joint ventures to deliver and deploy enterprise AI services. And when I say massive, I'm talking about a $10 billion starting valuation and a $4 billion investment for OpenAI and a $1.5 billion investment from Anthropic, with each of those endeavors bringing in just a who's who of financial and operational partners like Blackstone, Goldman Sachs, and more. These are companies whose fiercest battle is to be one-upping one another and pushing the pace of innovation ever forward. And yet they are taking time to frankly distract themselves on painful, boring day-to-day deployment issues because that's what it's going to take for even incredibly powerful technology, or perhaps especially incredible powerful technology to diffuse and actually have the impact that it could in the workplace. This maturation in our understanding of just how hard it is going to be to actually close the capability overhang and help AI do what it can do inside the enterprise is, I think, shifting people's timelines. And when it comes to the pain of disruption, timelines really matter. The world looks a lot different if there's a decade or two to adapt to changes as opposed to a year or two. When you're talking about a decade, things that wouldn't be possible in a year or two, like actual reskilling and redesign of different types of work and roles, start to look a lot more viable. Now, on top of the news that the biggest labs were willing to distract themselves with painful enterprise deployment issues, the other thing that I think has helped shifted people's narratives, which was the main subject of the weekly show last week, is the loss of the fantasy that somehow human-level AI is going to be massively cheaper than humans, at least in the short term. Last week was all about the shift in business model to usage-based instead of seat-based, which is a recognition that we are dealing with a world where there are far fewer tokens available than we would ideally consume. Which of course brings us to the second area and place in which the AI story is shifting, which is on Wall Street. Now, this started last week and has really continued this week where we are seeing both talk and behavior suggesting that Wall Street is not treating AI as a bubble that is inevitably going to pop as it might have been last fall. This week, both JPMorgan CEO Jamie Dimon and BlackRock CEO Larry Fink made comments affirming that the AI boom is real. Jamie Dimon said that he believed the trillion-dollar investment in data centers will make sense, his words, and Larry Fink went even farther saying not only is there not an AI bubble, but quote, "There is the opposite. We have supply shortages. Demand is growing much faster than anyone has anticipated. We have not begun exploring the opportunities of AI around the world." Now it's one thing for Wall Street leaders to say this, but markets are gonna do what markets are gonna do. And this week they seem to be on the boom train in a big way. Now last month we got the announcement of a 5-gigawatt deal between Anthropic and Google. We also knew that that deal had contributed to the $462 billion backlog for Google Cloud that was announced in last week's earnings. The information this week put a number on the amount contributed by that deal, however, reporting that it was worth $200 billion over 5 years. Now given that Google has also made an investment of up to $40 billion in Anthropic, This brought up for some, of course, the old arguments about whether this is all circular funding. And yet reaction to the Google-Anthropic deal has been very different than the last time these circular narratives were discussed. Google was already up 10% after the giant backlog was announced during earnings and spiked another 1.5% during the overnight session after the $200 billion number was reported. And in the back half of the week, Google not only didn't give up these gains, but firmed it up, adding another 0.5% as of the Thursday close. In other words, there doesn't appear to be any sign of the market worrying about Anthropic's ability to pay. Analysts see the strong and growing revenue allowing them to meet massive new commitments, and all of a sudden these numbers that seemed just so extraordinary before don't seem so out of reach now. Now I wonder how much of this is also driven by the fact that of the big frontier labs, Anthropic was comparatively conservative in their infrastructure deals and are paying the price now, having a race to catch up with better resource competitors in the form of OpenAI. Signal writes, "Just around 6 months ago, many people thought everything was a bubble. Too much compute, too much CapEx, and demand that couldn't possibly absorb the bailout. But it turns out the ceiling on demand for intelligence is literally nowhere in sight." Carmen Lee wrote, "Everyone's worried about a compute overbuild, but it's actually really hard to overbuild compute. Capital is the easy part. Money shows up fast. But money does not equal compute. You need GPUs, power, substations, colo, cooling, and operators." Each link has its own lead time. A capital bubble is a financing phenomenon. A compute bubble requires every physical bottleneck to clear at once. Now, the best place that I can show where this understanding of this particular story really came to bear this week is, of course, what will for sure go down as the biggest story of the week, which is the SpaceX team-up with Anthropic. On Wednesday, Anthropic and Elon Musk announced a new partnership where Anthropic will basically take over the entire capacity of the Colossus One data center. I talked about this extensively in yesterday's episode, but my thinking about this comes in a couple different parts. First of all, I think it just makes very obvious sense on an operational business level. xAI, now part of SpaceX, has struggled to produce a model that can keep up with the leaders, but has not so quietly built incredible capacity and compute and seemingly the ability to add more. Indeed, one of the things that we got with this Anthropic announcement was Anthropic throwing their weight behind the idea that data centers in space might not just be an Elon fever dream. So xAI, now a part of SpaceX, has a bunch of compute but no great models, while Anthropic of course has great models and real challenges with access to compute. On that level, the partnership makes complete sense. Why this is so exemplary though of the AI story shifting is mostly in the embodiment of Elon Musk himself being willing to shift his story vis-à-vis AI. While of course he has not given up on Grok, and is continuing to train future models in Colossus 2 with it. The fact that xAI is ceasing to exist as a separate company and is being completely folded into SpaceX, plus these big moves, suggest that Elon has gotten comfortable with the idea that his part of the story and his way to shape the future of the AI race might be more in infrastructure than in model development. To put a fine point on that, in follow-up reporting to the deal, the New York Times discussed TerraFab, which is Elon's chip manufacturing project in Texas. The new information comes from a legal filing in Grimes County where the project is based and says the project will cost at least $55 billion and possibly as much as $119 billion, which is way higher than the previous estimates of $20 to $25 billion. If completed, it will be by far the largest chip fab on the planet. Elon first announced TerraFab way back in March and people mostly brushed it off. Now people paid a little bit more attention when they added Intel as a partner in April, but you still saw folks like Nvidia analyst Tae Kim having a bit of skepticism. On TVPN, he said, "I'm not that optimistic. I mean, it's so hard to build that. It's almost like cooking where it takes a lot of trial and error accumulated over decades. It's not something you could just jump right in and do." And you're seeing this in the coverage now. The Anthropic deal adds a new level of credibility to the project. Back in March, even if you did believe that Elon could make the world's largest fab, it certainly didn't seem like he was going to have demand to justify it from just Tesla and Optimus. But now he has a basically unquenchable source of demand in the form of Anthropic. And so what you're seeing is not only Elon's story about himself shifting, but people remembering what was impressive about Elon in the first place. Peak Elon was scaling up Tesla production when he famously slept on the factory floor in 2018, and maybe the closest example we've had recently was him standing up the first Colossus data center in record time at the end of 2024. People are starting to remember that if there's one person who can pull off an insane construction and supply chain project like TeraFab, It's probably Elon Musk. One of the most important AI questions right now isn't who's using AI, it's who's using it well. KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising. The highest impact users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com/us/sophisticated. That's kpmg.com/us/sophisticated. Today's episode is brought to you by Granola. Granola is the AI notepad for people in back-to-back meetings. You've probably heard people raving about Granola. It's just one of those products that people love to talk about. I myself have been using Granola for well over a year now, and honestly, it's one of the tools that changed the way I work. Granola takes meeting notes for you without any intrusive bots joining your calls. During or after the call, you can chat with your notes, ask Granola to pull out action items, help you negotiate, write a follow-up email, or even coach you using recipes, which are pre-made prompts. Once you try it on a first meeting, it's hard to go without. Head to granola.ai/aidaily and use code AI Daily. New users get 100% off for the first 3 months. Again, that's granola.ai/aidaily. One thing I keep seeing in enterprise AI, companies hedging across every cloud, every model, every framework, or paying a GSI for a pilot that never ends. The team's actually shipping They've picked a lane and they move fast. That's one of the reasons I like today's sponsor, Robots and Pencils. They've gone all in on AWS. They're an advanced tier and AWS pattern partner, and they ship production AI coworkers in 45 days. That's led to them doing some of the more interesting work I've seen on AI coworkers. And by that, I'm not talking about chatbots. I'm talking about actual agentic systems that sit inside a business architecture and do real work. That kind of focus matters if you're an enterprise leader trying to get something real into production or an AWS rep trying to move a customer from interested to deployed. Request an AI briefing at robotsandpencils.com. One conversation with Robots and Pencils and you'll know. So coding agents are basically solved at this point. They're incredible at writing code. But here's the thing nobody talks about: coding is maybe a quarter of an engineer's actual day. The rest is standups, stakeholder updates, meeting prep, chasing context across 6 different tools. And it's not just engineers. Sales spends more time assembling proposals than selling. Finance is manually chasing subscription requests. Marketing finds out what shipped 2 weeks after it merged. ZenCoder just launched ZenFlow Work. It takes their orchestration engine, the same one already powering coding agents, and connects it to your daily tools. Jira, Gmail, Google Docs, Linear, Calendar, Notion. It runs goal-driven workflows that actually finish. Your standup brief is written before you sit down. Review cycle coming up? It pulls 6 months of tickets and writes the prep doc. Now you might be thinking, didn't OpenClaw try to do this? It did, but it has come with a whole host of security and functional issues which can take a huge amount of time to resolve. ZenCoder took a different approach: SOC 2 Type 2 certified, curated integrations, tighter security perimeter, enterprise-grade from day one, model agnostic, and works from Slack or Telegram. Try it at zenflow.free. Now staying in infrastructure, at the end of the week Nvidia announced a new partnership with Corning Glass, which manufactures the fiber optics which are the backbone of data center networking. Corning is close to a monopoly supplier for this type of glass, holding more than a 70% market share. The new deal will see Corning build 3 new facilities in Texas and North Carolina, adding 3,000 manufacturing jobs. Nvidia CEO Jensen Huang said, "We're going through the single largest infrastructure buildout in human history. Artificial intelligence is going to become fundamental infrastructure all over the world and surely here in the United States." He also used a language which we're starting to see come up a little bit more, when he said, "This is such an extraordinary opportunity because we can use these market dynamics to reinvest, revitalize American manufacturing for the first time in several generations." And here's the point that I want to make in the context of the AI story shifting. If you follow the logical chain from demand for tokens is insatiable, way more than our supply, to that demand represents a tiny sliver of overall total demand, because as we've discussed, A vanishingly small portion of the enterprise is actually fully onboarded into the agentic world. How much compute are we going to demand when it's not 5 to 20% of enterprises that are wired up and using agents, which by the way, I'm being generous, but 60 or 70%. The compute shortage starts to look monumental at that point. And of course that is exactly what investors are playing out, which is why they're back on board and excited once again about this data center buildout, but the next step after you recognize that the data center buildout that's been discussed so far is likely not a bubble in that it's going to have the token demand. And as we discussed last week, the usage-based business model to justify that token demand, then you start to realize that all of these new jobs that are being created around the data center boom are maybe not just some super temporary 2 to 5-year burst of construction activity. In other words, it's one thing when you have a job shortage that you need to quickly fill because you got a half decade of buildout. After which the buildout is done and the jobs boom is over. But that's not what we're talking about here. We are talking about a sustained, likely decades-long project to get access to the compute that we are going to need for the next phase of the global economy. And what that means is that that financial optimism that's showing up on Wall Street is going to pretty quickly diffuse into real enthusiasm in blue-collar sectors as the benefits of this buildout hit home in the immediate term. Now, as much as the data center conversation is fraught in terms of politics and has real issues and very legitimate issues from local citizens who have questions about data centers in their community, you're starting to see very clearly that when it comes to unions, especially construction unions, not only are they fully on board, they are leading the charge in trying to align the differences between data centers and communities so these projects can proceed. Craig Fuller tried to capture this on Twitter, writing, AI is driving an American manufacturing renaissance and will continue to do so in coming years. AI data center construction is the largest infrastructure investment in history, and most exciting, it's not coming from the federal government, but rather from private cash-flushed enterprises. The AI buildout requires massive capital spending, but not just on chips—on construction, power generation, equipment, etc. It's all infrastructure. AI data centers are massive. A 500-megawatt data center is the size of a mid-sized city airport and requires substantial concrete, steel, copper, fiber, piping, and huge cooling, transmission, and generators. It takes 30,000 truckloads to build one of these out, and that doesn't include the power plant that's required to run the thing. That is a massive project in itself. This will require massive capital investment for years to come. Thanks to tax incentives, almost all of these materials are being constructed in the United States, originating in the old manufacturing heartland, the Rust Belt, or South. Right now, I would say this view of the AI buildout as driving an American manufacturing renaissance is absolutely not dominant, but it is a bit ascendant. And that's what I mean by the story shifting. And when it comes to new products announced this week, I think a lot of them fit within the themes that we discussed both last week and this week in terms of a maturation of the product set and a real focus on solving the problems of these models in practice. In other words, there's a reason that we are in the harness engineering era. The raw model capability overhang is so immense that we kind of need the harnesses, in other words, the products that surround these models, to solve some key problems or else that overhang is never going to get closed. This week at the Code with Claude event, we got features focused on memory, features focused on solving human review, and more infrastructure tools around multi-agent orchestration. Cursor also added a skill around orchestration, introducing /orchestrate, which they call a skill that recursively spawns agents to tackle your most ambitious tasks with the Cursor SDK. From OpenAI this week, the big set of product announcements was around voice. Sam Altman tweeted, "People are really starting to use voice to interact with AI, especially when they have a lot of context to dump. GPT-Realtime-2 comes to the API today, and it's a pretty big step forward." We actually got 3 new voice models from OpenAI, all in the Realtime API. GPT-Realtime-2, which is their voice agent model, that can "think harder, take action, handle interruptions, and keep conversations flowing." GPT Realtime Translate, which is exactly like it sounds like, the ability to translate more than 70 input and 13 output languages, and GPT Realtime Whisper, which is their streaming audio transcription. Now I think the operative phrase from Sam Altman here is a lot of context to dump. This is 100% why voice matters so much right now and why I think it fits in this broader theme of just one by one going through the key issues of actually using agents in practice. One of the lessons that I have in every self-directed education program we do is to have people set up Whisper Flow or some version of it so they can start to talk to their computer instead of just type. And the reason for that is exactly what Sam says. You can dump context so much faster when you're talking than when you're typing. And now in the agent world, transmitting context from our brains alongside everywhere else into agents is kind of one of the key barriers to getting as much out of them as we could. Speaking of audio, ElevenLabs also announced that it had reached a half billion dollars in annualized revenue and added new investors including NVIDIA, BlackRock, Wellington, and Santander. And one more fundraising announcement that I found extremely as of the moment, past AIDB partner and sponsor Blitzy raised a couple hundred million bucks at a $1.4 billion valuation, becoming the latest enterprise AI unicorn. Now I'm old enough to remember when becoming a unicorn was a really big deal. But at this stage, it's almost more like, well, yeah, of course. It's just completely obvious that an enterprise-facing agent harness company that actually does its harness engineering well should be a unicorn in the current environment. Congrats to everyone at Blitzy, of course, and kudos to them for figuring out how important the harness was going to be before many others did. Now, as we round out, it's important to note that anytime we talk about narrative shifts, it's very easy to take it too far. One who wasn't looking for it could certainly see a lot of the same sort of discourse around AI this week that we've had in the past. Certainly the best example of this was the response of mainstream media to layoffs at Coinbase and then later in the week, layoffs at Cloudflare. Both of those companies prominently pointed to AI as a culprit, which most media outlets were completely comfortable running without question. And yet more so than we've seen in the past, there were a fair number of people who jumped in to look at the specifics of the companies before just accepting the AI layoff narrative at face value. Cloudflare, for example, was laying off 1,100 despite hiring 2,000 new people just a few months ago. Which of course makes the curious observer wonder if this wasn't more about correcting an overhiring. Meanwhile, I think Coinbase was even more transparent. The layoffs were announced before the quarterly earnings report, whereas anyone who has paid even the tiniest little bit of attention to markets and crypto for the last 6 months could have guessed what we were going to hear. Sure enough, in the last quarter, Coinbase transaction revenue fell 40% year over year. But as I said on Twitter, yeah, the layoffs were definitely about AI. The point is not to deny that AI had some role in these layoffs, We are in the midst of a shift in recalibration, and I think on average companies are going to be smaller in 5 years than they are today, even if they're producing more. What's different and where I think the story is shifting is with a bit of a rejection of the blind acceptance as AI is the default story behind any layoff that comes to the market's attention. Now, as we wrap up, let's talk about 1, what to watch for next week, and 2, what you should play with this weekend. For sure, the big story that I am watching for next week is what the White House does, if anything, around this notion that they might be vetting AI models before they get released. It is very clear that there are very different sides battling it out in this White House when it comes to what they should do around advanced AI models. At the beginning of the week, big outlets were reporting what would effectively amount to a reversal of position, having the White House be much more involved in AI than they had ever intended. But then by the end of the week, we got this piece from Politico suggesting that the White House was trying to distance itself from tighter AI regulation. One senior White House official told Politico, "There's one or two people who are very intent on government regulations, but they're sort of the minority of the bunch." So next week we'll be watching to see how this lands. It's very clear at this point that it is not a nothing story and is frankly a very dynamic and fluid situation. In terms of what to play around with this weekend, the answer is for sure, slash, goal. Investor, coder, lawyer, and researcher Dan Robinson summed it up like this: Codex just released the /goal feature. Tell Codex to set a goal and it will keep working on that goal until complete. Philip Kory, who works on Codex at OpenAI, called it "our take on the RALPH loop. Keep a goal alive across turns. Don't stop until it's achieved." A16z's Andrew Chen writes, "Trying /goal for the first time on Codex and it's obvious it's going to 10,000x token use. It's amazing though. I've had it working on a low-level eGPU+Mac device driver project overnight." that I have no business doing for the past 14 hours and it's still chipping away making progress with each iteration. Naturally unattended 24/7 LLM use will be several magnitudes bigger than me prompting actively over a normal workday. Alex Finn wrote, "You have to try the new /goal feature in Codex. It worked for over an hour and built me an entire complex extraction shooter video game. You give it a goal, then it works endlessly until the goal is complete. It's like a Ralph loop, can run for days. If you enable the imagegen skill before you run the goal, it will even generate all the assets for your game autonomously." A couple days later, he followed up: "The biggest advancement in AI coding this year has been /goal and it isn't even close. It allows your AI agent to quite literally work for days without stopping. You give it a mission, it works until the mission is complete." However, he says, "/goal is useless if you don't use it properly. You need a good prompt for it. I've found basically any prompt I handwrite after /goal is never good enough. It produces results that might as well have been a normal prompt." Meta-prompting is the answer. Go to any AI that has context around the project you're working on. Say, "I'm working with Codex and I want to use their new /goal feature. Please research their /goal feature then take a look at our project and give me 3 options for how we could use /goal to be maximally productive. Then give me a highly detailed /goal prompt for each." Take one of the prompts, then go to the Codex CLI and type /goal then give the new prompt. I 100% guarantee the AI does better work than you've ever seen before. Now for what it's worth, this is exactly what I did. I went to the new GPT-5.5 model, said, go research the new /goal feature and tell me what types of projects it's well suited to. And then from there, slowly whittled it down to the projects that I am working on across the AIDB suite, deciding ultimately to have it work on a forthcoming thing called AIDB for Teams that basically takes the daily episodes and turns them into chunked insights that are designed for actual knowledge workers inside companies that allow them to get the highlights without having to consume the entire 20 or 30 minutes. To give you a sense of how /goal fits, when I asked if that would be a good project for /goal, 55 responded, "Yes, this is a real /goal-shaped idea, but only after you separate two things. Building the system is a normal Codex project, but running the system every day against a new episode can become a /goal project. The key is, can the objective be made persistent, inspectable, and verifiable?" And the idea is that the daily episode translation is exactly that. It needs a common set of formats and outputs, strong consistent asset generation, and more. So I will be doing that experiment this weekend, and I will be excited to hear if any of you do as well. For now though, that is gonna do it for today's AI Daily Brief. Again, as always, if you have any feedback on this new weekly format, please let me know. For now, just a big thank you for listening or watching, and until next time, peace.