How Harness-as-a-Service Will Change Agents

Today on the AI Daily Brief, the emergence of harness as a service what it means for the agentic era, and before that in the headlines, a big tech AI earnings blow out. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Ah, alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors KPMG, Blitzy, Granola, and Section. To get an ad-free version of the show, go to patreon.com/airdailybrief or you can subscribe it on podcasts. If you want to learn more about sponsoring the show, send us a note at [email protected]. And lastly, if you haven't yet go check out the new Agent OS program. It's a tool agnostic adaptable system for building an agentic operating system. And I have a feeling after you listen to today's harness as a service episode that you will want to dig in even more, you can find that off of the main site, aideilybrief.ai. Today is one of those rare days where the headlines are all around the same theme and that theme is of course big tech earnings. And to not bear the lead, let's go over to Shabelur who writes, "Hard to take the AI bubble argument seriously when some of the largest companies on earth are still putting up these growth numbers. Google Cloud plus 63% year over year, Microsoft Azure plus 40% year over year, MetaRevenue plus 33% year over year, AWS plus 28% year over year." We're going to go through all of these and talk about the winners and losers and what it means for the market's assessment of AI overall. Google was the clear winner on big tech earnings night delivering huge beats across the board. They reported 22% top line revenue growth but as I just mentioned the big numbers were in their AI related businesses. Google Cloud has experienced 63% revenue growth over the past year. They also reported a $460 billion backlog in new orders up from 240 at the end of Q4. Their new deal with Anthropic contributes a decent chunk of that growth but it still shows that GPU demand is off the charts. Analysts shows if Carlson posted the chart of Google's cloud backlog going exponential and commented, "This is so crazy it literally looks fake." Gemini Growth was similarly strong. Google reported a 40% surge in paid enterprise customers quarter over quarter meaning maybe I'm going to have to eat my hat rating Google as low as I did on enterprise in the AI lab power rankings. In addition, Google's infrastructure is now processing 16 billion tokens a minute up 60% quarter over quarter. Even search which is tangentially related to some of Google's AI changes is experiencing a boom with search revenue up 19% year over year. Google is also maintaining very healthy profit margins hitting 62.6 billion in net income for an 81% year over year gain. CEO Sundar Pachai told analysts that AI is now the largest tailwind for cloud commenting, our enterprise AI solutions have become our primary growth driver for cloud for the first time in Q1. He added though, "We are compute constrained in the near term. Our cloud revenue would have been higher if we were able to meet the demand." CapEx targets meanwhile remained pretty anchored. Google slid up their forecast slightly increasing this year's guide from a range of 175 to a hundred and eighty five billion to a range of 180 to 190 billion. Although right now they're not coming close to hitting that target. They only recorded 35.7 billion in CapEx spending for Q1 which annualizes out to 140 billion in change. Now Google might be patting the back end of the year in expectations of rising costs, but because this came in slightly below analyst forecast the market interpreted it as capital discipline. Google was up 7% in overnight trading. Chubby meanwhile argued that Google has cracked the AI monetization thesis. Although the absurdly high cloud growth was the loudest signal, Chubby added, "The real story is search." The prevailing narrative was that AI would cannibalize Google's core business, people get answers from chatbots, stop googling. The opposite is happening. Search ad revenue grew 19%, queries hit an all time high. Google turned the biggest existential threat to its business into a growth accelerator. Amazon's earnings were also extremely solid, suggesting a return to form on cloud. Topline revenue growth came in at 17% year over year with net profit up 77%. Amazon attributed the growth in profit to pre-tax income from their anthropic investment, so this number could say more about anthropic than it does about Amazon. AWS revenue was up 28% compared to last year, with Amazon's dominant cloud unit coming out of a major slowdown after achieving a low of 12% growth in 2023. Growth has been accelerating every quarter since and is now back at a healthy level. The number beat analyst forecast and would have been viewed very solid if not for the direct comparison to Google Cloud's incredible 63%. With the extent that there was an issue, it was in Amazon's costly AI buildout. Amazon said they had added more server capacity than any other company in 2025 and planned to accelerate construction even more this year. Their capital spending for Q1 was 43.2 billion, putting them slightly behind their $200 billion target pace for the full year, but that remains a 60% jump from last year and the largest overall number between the four big tech hyperscalers. Amazon is now essentially spending every dollar they make on the AI buildout, driving free cash flows for the quarter up from almost 26 billion this time last year, down to 1.2 billion this quarter. CEO Andy Jassy dismissed concerns, stating the company would keep spending to keep up with demand. He added that he was confident that additional spending would translate into strong profits because most of the new supplies are already spoken for. And of course with OpenAI now partnered with AWS alongside anthropic, that is probably a fairly safe bet. The Wall Street Journal called the AWS expansion prescient, writing, "The growing demand for chatbots and other AI powered tools is outpacing the supply of chips and storage, causing outages and surging prices." Jassy also noted that Amazon's in-house chips have been well received, commenting, "We have such demand right now for training from various companies who will consume as much as we make. I expect over time there's a good chance we're going to sell racks over the coming years. We have to decide how much we're going to allocate to the existing demand and how much we're going to save to sell as racks." He commented that if their chips were a standalone business in actually booking revenue from AWS, they would be sitting at $50 billion in ARR. Jassy boasted, "As best as we can tell, our custom silicon business is now one of the top three data center chip businesses in the world, and the speed at which we've gotten here is extraordinary." Now, one of the big themes over the past year has been Amazon reasserting their dominance as the largest cloud provider. There's been far less emphasis on training their own models and releasing AI products. Instead, they're forming strong strategic partnerships and putting themselves in prime position as they pick and shovels AI infrastructure play. Jassy reinforced that strategic pivot during the earnings call, commenting, "We're in the middle of some of the biggest inflections of our lifetime, we're well positioned to lead, and I'm very optimistic about what's ahead for our customers and for Amazon." Still the market had a bit of a lukewarm response. The stock was down initially but ended the overnight session up 2.6%. Many attributed this though to the market already boosting Amazon on news of the open AI deal earlier in the week. The logic is that with the stock already elevated, it would have taken a big upside surprise to push it even further. Most of the commentary centered around AWS writing the ship with a little help from the gigantic compute crunch. She below again wrote, "AWS is now a $152 billion ARR business growing 28% per year that marks its fastest growth in nearly four years." Road signal, people underestimate how big of a deal it is that open AI models are now on bedrock. I met so many companies that defaulted to anthropic and cloud because they were already on bedrock and for a long time that was basically the path of least resistance. This is huge for open AI model accessibility. Moving on to Microsoft, the company has been sort of in this middle space of the three big cloud giants. They didn't commit to aggressive growth like Google, instead opting for a more conservative AI CapEx strategy. They also have scale and in-compancy disadvantage when compared to AWS. Most of what they did have was exclusive distribution rights to open AI models and that's now gone. At the same time, they're one of the few companies selling tokens during a token drought and that counts for a whole lot. Big story from earnings was 39% growth for Azure. That was right in line with expectations in just a 1% point acceleration from the previous quarter. Top line revenue was a slight beat coming in at 82.9 billion for an 18% year over your gain. In terms of forecasts, CFO, Amy Hood projected this growth rate to continue at a steady pace anticipating Azure's 40% growth rate to continue for Q2. CapEx guidance was lifted by 25 billion to reach 190 billion for the year. Hood attributed the entire increase to higher component prices rather than adding more data center projects. On co-pilot CEO, Satya Nadella said that Microsoft now has 20 million paid seats for the enterprise add-on up from 15 million in January. Microsoft expects that growth to continue into the third quarter. It appears that utilization is also up with Nadella commenting. Weekly engagement is now at the same level as outlook as more and more users make co-pilot a habit. Responding to questions about losing exclusive access to open AI models, Nadella downplayed concerns stating, we have a frontier model royalty free with all the IP rights that we will have access to all the way to 32 and we fully plan to exploit it. Over the past few months, the market has lumped Microsoft in with software stocks and included it in the SaaS apocalypse with the stock down more than 10% so far. Unfortunately for them, last night's earnings didn't seem to do anything to change that viewpoint with the stock whipsawing up and down throughout the overnight session but ending the night basically flat. Ultimately nothing disastrous or spectacular came out of the earnings call. They continued to add GPUs to their fleet at a tremendous pace but a pace that feels conservative compared to Google's numbers. It's a big positive to see co-pilot growing but the negative take is that 20 million is still a drop in the ocean compared to the roughly 320 million paid seats for Office 365. After earnings were released but prior to the investor call, Gene Munster of Deepwater Management wrote, "Microsoft is in a tight spot because they can't shake the negative narrative. Stock is down 1% on solid numbers. This past month Microsoft is up 17% in line with the NASDAQ. My take, they need to make a statement on the call tonight and make it easy for investors to believe that they have AI powered products beyond Azure that customers must have. Co-pilot has been a miss." Nadella ultimately didn't deliver that big statement and so the company continues to perform like a perfectly average tech stock. Not a bad place to be during the AI boom but Microsoft is starting to fall behind their peers who are making bigger moves. Now on the meta front, the past year has been all about rapidly expanding capex. Beginning in 2024, Mark Zuckerberg seemed to make the decision that he would continue accelerating data center spending regardless of investor concerns. Meta now finds it. themselves with the largest market cap adjusted CapEx commitments among the tech giants. Over the past few quarters, the narrative was that revenue growth was accelerating fast enough to justify the AI spend. And that Zuckerberg had earned the right to spend big on data centers by delivering solid returns. Analysts were linking increasing advertising revenue directly to AI optimizations in Metazad platform. Heading into Wednesday night's earnings, the balance between revenue growth and CapEx remained a central question. The financial reports showed that Meta had delivered another record quarter. Quarterly revenue came in at 56.3 billion up 33% over the past year. Both total revenue and net income beat analyst forecast by a fairly significant margin. On the other side of the ledger, Meta hiked capital expenses once again. They increased this year's forecast from 135 billion to 145 billion. Notably, this wasn't because Meta was tacking on another data center or two. The CapEx hike was instead attributed to expectations for higher component pricing and additional data center costs. Meta also disclosed a quarter over quarter decrease in daily active people, which is Meta's in-house metric for user activity. The decline was small and Meta blamed internet disruptions in Iran alongside a restriction on WhatsApp and Russia. Still, it's the first time that Meta's user metrics have decreased since they first began reporting it in 2019. Advocating for his data center build out, Mark Zuckerberg told investors, "Every sign that we're seeing in our own work and across the industry gives us confidence in this investment, that said we are very focused on increasing the efficiency of our investments." Meta CFO Susan Lee noted that Meta is suffering from the same compute shortages everyone else, even though most of their AI computers deployed towards internal and infrastructure use rather than customer use. Lee said, "Our experience so far has been that we have underestimated our compute needs, even as we have been ramping capacity significantly. So our expectation is that compute will become even more central to the business going forward." In any other year, this would have been a blowout earnings result. A 33% revenue growth rate is the highest they've achieved since 2021 and up substantially from 23.8% for the previous quarter. And yet, Meta was the biggest loser from the big tech earnings. The market still hates their CapEx spend and send the stock plummeting by 5% in overnight trading. Jim Kramer summed up the market opinion when he said, "Metage did not offer enough reasons to spend the way that other companies did. They just told us that they could do better with it, hence the decline tonight." If there is one big takeaway from tech earnings, it's that the AI boom is in full effect. Everything from memory chip foundries to data center construction is running at 100% in a frankly vain attempt to keep up with the endless demand for tokens. This is categorically different to last year. When many analysts still had questions and AI optimists were betting on a forthcoming boom, that boom is now clearly here and it's showing up in big tech's growth numbers. There are still of course reasonable questions around business models and profitability, as well as who the biggest winners will be. But AI demand is unquestionable at this point and appears to still be accelerating. For now that that's going to do it for today's headlines, next up the main episode. 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. Want to accelerate enterprise software development velocity by 5x? You need Blitzy, the only autonomous software development platform built for enterprise code bases. Your engineers to find the project, a new feature, refactor or greenfield build. Blitzy agents first ingest and map your entire code base, then the platform generates a bespoke agent action plan for your team to review and approve. Once approved, Blitzy gets to work autonomously generating hundreds of thousands of lines of validated end-to-end tested code. More than 80% of the work completed in a single run. Blitzy is not generating code, it's developing software at the speed of compute. Your engineers review, refine and ship. This is how Fortune 500 companies are compressing multi-month projects into a single sprint, accelerating engineering velocity by 5x. Experience Blitzy first hand at Blitzy.com. That's BLI-Tzy.com. Today's episode is brought to you by Grenola. Grenola is the AI notepad for people in back-to-back meetings. You've probably heard people raving about Grenola, it's just one of those products that people love to talk about. I myself have been using Grenola for well over a year now and honestly, it's one of the tools that changed the way I work. Grenola takes meeting notes for you without any intrusive bots joining your calls. During or after the call, you can chat with your notes, ask Grenola 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 Grenola.ai/aiDaily and use code AIDaily. New users get 100% off for the first three months. Again, that's Grenola.ai/aiDaily. Here's a harsh truth. Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized. Half of companies have AI tools, but only 12% use them for business value. Most employees are still using AI to summarize meeting notes. If you're the one responsible for AI adoption at your company, you need section. Section is a platform that helps you manage AI transformation across your entire organization. It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value. The result? You go from rolling out tools to driving measurable AI value. Your employees move from meeting summaries to solving actual business problems, and you can prove the ROI. Stop guessing if your AI investment is working. Check out section at sectionai.com, that's SEC T-I-O-N-A-I-D-D-D-D. Welcome back to the AI Daily Brief. Nominally our topic today is Kurser's new cursor SDK. It's a platform where as Kurser's Lee Robinson put it, you can build local hackable agents with any model or ship products on top of managed cloud agents. The cursor SDK handles the harness, sandboxing computer use demo videos, GitHub integration, etc. We're going to talk about the cursor SDK, what it does, what people are starting to build with it, but I want to make an argument that it's actually part of a broader phenomenon. In the past few weeks alone, we've seen OpenAI make a big update to their agents SDK, we've seen Anthropic released cloud managed agents. Microsoft released hosted agents in Foundry, with CEO Satya Nadella writing, "Every agent will need its own computer, and with new hosted agents in Foundry, every agent gets its own dedicated enterprise-grade sandbox with a durable state, built in identity and governance and support for any harness or framework." Kurser SDK is not the same as all these products, but they are all playing around in similar space. And for the sake of simplicity, I'm going to propose a new name for this category, Harness as a service. A new infrastructure category where companies sell access to their agent runtime, which is the engine that turns an LLM into something that can actually do work, the same way AWS sells access to compute and striped sales access to payment rails. So before we can go to the next level in explaining this, let's take a step back and redefine Harness itself. You might have watched or listened to my show about Harness engineering a couple weeks ago. It's a term that has been on the rise all year. Okshay on Twitter actually did a nice little summary. He argued that the agent landscape has evolved in three phases, writing that the biggest shift in AI agents had nothing to do with making models smarter, but was instead about making the environment around them smarter. In Phase 1, the weights phase he writes, "Everything was about the model itself, bigger models, more data, better training." Scaling laws told us that progress equals more parameters. RLHF and fine tuning shaped behavior, and if you wanted a better agent, you trained a better model. In Phase 2, the context phase, however, there was a new realization. You don't always need to change the model. You can change with the model Cs. That got us prompt engineering, few shot examples, chain of thought, rag. Suddenly, he writes the same frozen model could behave completely differently based on what you put in front of it. Developers stopped fine tuning and started iterating on prompts and retrieval pipelines instead. But he continues context windows or finite. Long prompts get noisy, models attend unevenly, and every new session starts fresh with zero memory of what happened before. Which gets us to the third phase, the harness engineering phase. Okshay writes, "This is where we are now, and the shift is fundamental. The question changed from what should we tell the model to what environment should the model operate in. The model is no longer the sole location of intelligence. It sits inside a harness that includes persistent memory, reusable skills, standardized protocols like MCP and A2A, executions handboxes, approval gates, and observability layers. The model stays the same. What changes is the task it's being asked to solve. One concrete example he gives is a coding agent asked to implement a feature run tests and open a PR. Without a harness, he writes the model must keep repo structure, project conventions, workflow state, and tool interactions all inside a fragile prompt. With the harness, persistent memory supplies context, skill files and code conventions, protocolized interfaces and forced correct schemas, and the runtime sequences steps and handles failures. Say model completely different reliability. Importantly, he writes, each phase didn't replace the previous one it layered on top. Wait, still matter. Context engineering still matters, but the center of gravity has moved outward. Now, I think intuitively, people have experienced this to some extent. This agentic explosion that we've experienced in 2026 was not just the byproduct of Opus 45 and GPT-5 too, although those model updates were integral to it. It was the combination of those models with their harnesses, Cloud Code, which was the default harness for agent builders throughout 2025, and more recently, of course, OpenAI's Codex. The conversation has matured at the point where in a recent interview with Ben Thompson, Sam Altman was asked, "How important is the harness, the runtime around the model, the tools, the state, to making agents actually work?" Sam said, "Hard to overstate how critical it is. I no longer think of the harness in the model as these entirely separable things. Like, my experience of using these, I am very aware of the fact that I don't always know when I fire something off in Codex and it does an amazing thing for me. I don't know how much credit." Ben Thompson Philzin was that the model is amazing or the harness is amazing, to which Altman responds, "Yeah, exactly." So we now have these two very different vectors of increasing AI capability. There's the underlying models and the changes there, but then there's also improvements in the harness that surround them. Of course, another part of what has made this year feel so different and what has empowered so many different people is that we got an open harness, even though that's weren't what most people were calling it at the time, that actually allowed us to build the structure around agents to have them do the things that we'd always imagine agents being able to do. I'm talking, of course, about OpenClaw. And yet, OpenClaw was not plug-and-play. You had to do everything from picking the model to write the system prompt to define the tool, to wire the agent loop by either part that decides what to do next, that dispatches tools handles results decides when to stop, you had to manage context, you had to handle errors. The user building their OpenClaw had to orchestrate sub-agents when they needed parallel work. They had to figure out how to store state between runs, they had to figure out where to deploy it, how to monitor it, and if something broke, you fixed it. If you wanted a new capability, you built it. Every layer of the stack was yours to assemble, configure, and maintain. I don't think it's a totally inappropriate analogy to look at this almost like the hobbyist era of computing. In a recent post on LinkedIn, Anders Carlson wrote about the forgotten era in computing. In the post, Anders talks about how they started with a computer in their school, a little mini computer called an Alpha LSI, but quickly decided that they wanted a computer for their self. They continue. CompuKit UK101. This was a kit in the true sense of the word. Not just a pre-produced circuit board that you put in a case, but an unpopulated circuit board, some chips and other components, and a very rudimentary assembly manual. You needed a soldering iron, some basic tools, and a ton of patience to build this thing. This might be hard to grasp for people using computers these days, but for a few years in the 1970s, just after Ed Roberts had released the Altar 8800, and before the Apple II, and that generation of computers came around, there was a short era of these truly hobbyist computing. In other words, there was this period. Believe it or not, where the way that a lot of people were interacting with computers was having to build and assemble them themselves. You got a kit in the mail and you had to put it together. Now obviously this period didn't last long, and the number of people who were willing to do this was pretty small. It was the opportunity that companies like Apple with their Apple II seized upon in order to truly get us into the PC era of today. Now at this point, you might be already following the analogy that I'm about to make. But effectively, the open-cloth era of harnesses, you know, in the ancient days two and a half months ago, was an era where there were no pre-built harnesses you could rent. You could either use the basic harness plus model integration of codex and cloud code, or you could customize everything in open-cloth. What we're getting now, with what I'm calling harness as a service, is structurally different. With these tools, the agent loop is pre-built. The tool dispatch is pre-built. Sandboxing is pre-built. Streaming, error handling, context compression, all pre-built and tuned by teams whose full-time job is making those layers excellent. You bring three things. Which model you want, what tools the agent has access to, and what task you're handing it. Everything underneath is handled. This is obviously not a shift in scale, but a shift in kind. And just like the PC era didn't destroy or undermine all those hobbyists who liked having control over all the components, it was an incredible democratizing force that made it possible for people who never would have built their own PC to actually use a computer. The productivity revolution of the 1990s happened because users got Dell desktops not because more people learned to assemble motherboards. That is the promise of this new harness as a service phase. Now the analogy isn't perfect, because the actual democratizing force when it comes to agents won't really for most people be building their own agents. It'll be using some set of prefabbed agents as a service that they have a very small number of knobs to customize, and the applications that are powered by agents that are more customer-facing. What's weird and interesting and different is that while on the surface of it, tools like managed agents and the cursor SDK seem like they're only for developers. Because we now live in the world where agents handle the coding and can support you around the infrastructure build, the audience of people who can build with something like the cursor SDK has expanded dramatically. We've got this entire new category of builders of whom I'm an example who are not developers, but for whom something like the cursor SDK will matter. Let's give an example of something that someone built with this. Jack Driscoll showed off a demo after building with the SDK for a few days on pre-release. The agent is effectively a cursor agent embedded directly into Gmail. From within his Gmail inbox, he can share an email into chat, have that agent read the email thread into context, go off and do something, i.e. edit some code, fix a problem, etc. and stream the results right back into the chat window. When someone asks why this is different from just building with any other SDK, Driscoll commented, "The biggest difference in my opinion is that cursor SDK isn't just call an LLM with tools. It's exposing the same coding agent runtime cursor already uses. Rebo context, edit search, terminal workflow, streaming status, model choice, and local hosted execution. For this demo, Gmail and chat is just the intake and collaboration layer. Cursor SDK is the part that can actually go operate on a code base like a cursor agent would. In short, the harness. In other words, Jack has taken advantage of this new harness as a service to build exactly the agent wants in exactly the environment he wants doing the exact thing he wants. Now what's interesting is alongside this announcement, we're seeing how much harnesses can change the performance of the underlying models. A new report from Endor Labs found that GPT-5-5 operating within cursor's harness set a new record on their security correctness benchmark. The benchmark test code for both functionality and security. The cursor plus GPT-5-5 combination achieved a score of 23.5% on the security section, which narrowly beat out the previous leader, which was cursor in Opus-4-7, with a score of 22.9%. Both of these scores were a few percentage points higher than Opus-4-7 in its native cloud code harness and GPT-5-5 in its native codex harness. The functionality test was arguably even more stark. Opus jumped by almost 4 percentage points from 87.2 to 91.1% by switching to cursor, while GPT-5.5 went from 61.5% to 87.2% by switching to cursor. Endor Labs wrote, "The key takeaway. Same model, same week, two harnesses, two different functional results." Alex Vultov from the Thursday AI podcast confirmed that he found similar results with an entirely different setup. Testing the models on coding benchmark Wolf Bench AI, he found that cursor's harness had the strongest performance for GPT-5-5 and was on par with cloud code when running Opus-4-7. With the SDK just launching yesterday, we're already seeing a ton of people building MbPs for agetic products with a new platform. Teas have very built a bug catching agent that can work on its production codebase and can see how the app is performing on its own browser window. This kind of agent could be a big step to figuring out how to create fully autonomous coding agents given that human code verification is currently a massive bottleneck. Hevery wrote, "It's mainly about closing the feedback loop. Right now, agents write code and hope it works. They can run tests, but tests don't catch everything, especially UI behavior, integration issues or flows that depend on real browser state. He explained that being able to actually see the app as a huge change, allowing for faster iteration, more confident agent handoffs, and far less verification work for humans. Robert Boucherie used the SDK to embed a cursor agent in a Chrome plugin. The tool is designed for IT triage, helping non-technical users dump code from the browser into a ticket instead of just describing the bug and hoping for the best. Obviously, these use cases are just scratching the surface, but they demonstrate the power and potential of freeing cursor agents from their IDE container while retaining their runtime environment. Basically, an entirely new universe of agentic apps just opened up, now the developers don't need to start from scratch. Now, what's very clear when you watch this space closely is that we are going to see massive amounts of iteration and exploration around harnesses. What's interesting to me about harness as a service is the extent to which those innovations don't just end up producing better harnesses that are embedded in the tools we use, but actually allow people to build whatever it is that they want to build, leveraging those harness innovations at a more fundamental level. If you want to check this out for yourself, but you are not a developer, what I would suggest is go to the cursor announcement, click on the link to their GitHub cookbook, drop it into either cloud or chat GBT, especially if it's got context about your particular projects, and ask it to give you some ideas for how this new harness as a service product could change the way that you build or think about something. Then of course, when you realize that it unlocks something that you've been wanting to do forever, I give you full permission to call and sick on Friday and just dive all the way in. For now that's going to do it for today's AI Daily Brief, appreciate you listening or watching as always and until next time, peace!