# The Rise of Agentic Coding and AI Infrastructure Constraints

**Podcast:** The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis
**Published:** 2026-04-15

## Transcript

Today on the AI Daily Brief, Vibecoding gets an upgrade.
Before that in the headlines, the new models that we are anticipating this week and the ones we've already got.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
Alright, friends, quick announcements before we dive in.
First of all, thank you to today's sponsors, KPMG, Mercury, Granola, and Blitzy.
To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts.
If you are interested in sponsoring the show, send us a note at sponsors at AIDAybrief.ai.
And of course, while you are at AI DailyBrief.ai, you can find out everything about the show and the larger ecosystem.
Two things I would suggest you check out.
First of all, we have brought back the newsletter if you are ever looking for the links on the show.
They're all going to be there.
Second, I'd also love you to join our AI operators community.
There's going to be a lot more going on over there soon.
With that out of the way, though, let's talk some new models.
The most fun weeks around here are, of course, the weeks when we get big new products or big new model releases.
As you'll hear in the main episode, we've got a bunch of new upgrades around agentic coding that are coming to places like Lovable and Claude Code, but according to the information, we are also getting a new anthropic model this week, although it is not the mythos model that everyone was talking about.
Instead, sources told the information that Opus 4.7 could be released by the end of this week.
And in addition to the new model, Anthropic is preparing a tool for designing presentations, websites, and apps.
Now, this is a similar segment to that targeted by Google with Google Stitch.
Theoretically, alongside AI native design tools like Gamma, this would also put Anthropic into direct competition with incumbents like Figma and Adobe.
Now, when it comes to Opus 4.7, rumors are fairly thin.
Obviously, just based on the name alone, most expected to be an incremental improvement over the previous version.
However, the reporting did trigger some rumors out of the other leading lab.
OpenAI leaker IRule the World wrote, Thursday is big, huge, delicious.
Looks like someone leaked to Sam that Opus 4.7 is coming.
Launch dates moved to Thursday, and as I broke yesterday, Opus is coming.
No more vague posts.
Spud and Opus 4.7 will launch on Thursday.
If this is not the case, I will no longer mention release dates.
Now, I do have to say iRule the World has not necessarily had the best track record lately, but there is a noticeable uptick in OpenAI vague posting, so who knows.
When it comes to Opus 4.7, Anthropic's in kind of a tough position.
They've let everyone know that they have this big, exciting new model in Mythos, and no matter what, it would be hard to release a lesser model when people knew that that other one was available to others who weren't them.
They're also dealing with increased narrative profession.
Shravan the Boyant Man on Twitter wrote, Opus 4.7 will be no better than 4.6 or 4.5, but people will applaud 4.7 because Anthropic conditioned people to think 4.6 sucks by nerfing it over the last four to five weeks.
Now one model that has been released is GPT-5.4 Cyber.
It's a variant model designed to address the increasing interest in cybersecurity on the back of the Mythos preview.
Writes OpenAI, the progressive use of AI accelerates defenders, those responsible for keeping systems, data, and users safe, enabling them to find and fix problems faster in the digital infrastructure everyone relies on.
Similarly, AI is being used by attackers looking to cause harm.
GPT-54 Cyber then is a variant of OpenAI's flagship model, designed to be more permissive around cybersecurity functions, which are often blocked by safety guardrails.
The model is only available to verified users, but as part of its introduction, OpenAI is expanding their trusted Access for Cyber program.
This program ensures that dangerously unrestrained models are only available to known cybersecurity professionals, with OpenAI saying that the access program has now expanded to thousands of verified individuals as well as hundreds of teams.
In their blog post, OpenAI listed democratized access as one of their main goals, attempting to, quote, make these tools as widely available as possible while preventing misuse.
To that end, OpenAI has set up a clear objective standard around qualification for the program, which includes identification and verification of profession.
And although this seems small, you can absolutely see the emergent strands of some new brand positioning coming together.
OpenAI appears to want to increasingly call out Anthropic for concentrating power in the hands of a very small number of people and firms, and so OpenAI is trying to position itself as the opposite of that.
They write, We aim to make advanced defensive capabilities available to legitimate actors large and small.
As part of the release, OpenAI is also rolling out a custom harness called Codex Security, which is of course designed to enable cybersecurity work.
Now, this new narrative that I'm mentioning is clearly not a slam dunk so far.
The X thread announcing the release was filled with people complaining about the conflict between claiming to democratize access while still gating the release to vetted cybersecurity professionals.
Some of those said professionals even reported missing out on access without understanding why.
Now, speaking of things that people are going to complain about, Anthropic is switching to usage-based pricing for their heaviest enterprise users.
The information reports the pricing change has already gone into effect, requiring enterprises to pay for usage on top of a flat $20 a seat charge.
Frederick Flipson, a software licensing consultant at Redress Compliance, said these customers are already paying $200 a month for each user, but were generally still receiving a discount compared to token usage.
He estimates that Anthropic's leading customers could double or even triple their costs under the usage pricing model.
Anthropic said that the pricing change wouldn't impact companies with fewer than 150 users, and some customers wouldn't see a price change until they renew their contracts.
They argued that the pricing structure is largely about removing usage limits as a problem for heavy enterprise users.
Now, this shift, of course, comes amid a drastic shortage in AI compute.
Anthropic has been hurriedly making deals to expand their GPU fleet, but the problem is industry-wide.
The Wall Street Journal reports that market pricing for GPU rentals are up 48% over the past two months.
JJ Cardwell, the CEO of cloud infrastructure company Volter, said, there's a massive capacity crunch that's unlike anything I've seen in more than five years I've been running this business.
He also added that not much can be done, saying, the question is why don't we just deploy more gear?
The lead times are too long.
Data center build times are long.
The power that's available through 2026 is already all spoken for.
Meanwhile, on the enterprise side, AI is busting budgets.
During a recent interview, Uber CTO Nepalinaga said that the rise of Cloud Claude Coate has already eaten his entire AI budget a few months into the year.
He commented, I'm back to the drawing board because the budget I thought I would need is blown away already.
Uber has taken the common approach of encouraging AI adoption with leaderboards for engineers to track who is using the most AI, and Naga said that they're seeing huge value and doesn't plan to rein in AI spending.
He said that 11% of Uber's back end code is now written by AI agents up from a fraction of a percent at the beginning of the year.
Now, obviously, this is one of the biggest substores that we are tracking on this show.
The cost profile of AI obviously has fairly dramatic implications on just how much it even has the capability to replace workers in short order.
Meanwhile, over on Wall Street, the SAS pocalypse seems to be back on at the reports around Anthropic's new design tool seem to have reignited Wall Street's fear.
Yesterday, Adobe dropped 1.8%, Wix fell 1.7%, and Figma led the drawdown at a 4% loss.
The dip for design tools happened alongside an otherwise hot market, with the Nasdaq rising 2% on Tuesday on hopes the Iran War has reached a steady state.
Summarizing the state of things, Bugo Capital writes, The market is just a guy staring at two screens.
One has Truth Social, the other is Anthropic's blog.
In front of him are five buttons that say software, semis, European defense, energy, and gold.
Trump or Anthropic Post, and he hits a button to make those stocks move plus 5% or minus 5%.
Pretty pretty much.
Lastly today, an update on the policy side.
My home state of Maine has banned data center construction for the next 18 months in the first statewide moratorium.
On Tuesday night, main lawmakers approved a bill banning the construction of new data centers until November 2027.
The legislation would apply to any data centers over 20 megawatts, which is of course a trifling amount in the era of gigawatt scale deployments.
It also establishes a mechanism to study the impact of data centers on the electric grid.
It's unclear whether Governor Janet Mills will sign the legislation into law.
Her office didn't indicate her stance, and Mills earlier lobbied for an exemption to allow a data center to be built in an abandoned paper mill.
Still, this is the first moratorium to pass through a state legislature and could be far from the last.
Twelve other states have bills on the agenda, ranging from Deep Blue New York to bright red Oklahoma.
Even Virginia, which houses the largest concentration of data centers on Earth, is looking to remove tax breaks after a moratorium bill failed.
The sponsor of the main bill, Melanie Sachs, who also chairs the Energy Committee said, this is not a Walmart.
Having a projected load of even two or three data centers can really impact the state as a whole.
She said that the moratorium is about, quote, stepping back and saying, if this is going to come, what can we do to really make it beneficial?
Now the business community in Maine is meanwhile up in arms about the legislature blocking desperately needed investment in rural parts of the state.
Patrick Woodcock, the chief executive of the Maine State Chamber of Commerce, argued that the ban is largely performative but could cause real harm.
He said, it's not as if we don't have a review process for these projects to begin with.
There are real safeguards in place for ratepayers and the environment.
Lots more to say about this one, but for now, the moratorium era has begun.
That, however, is going to do it for today's AI Daily Brief Headlines edition.
Next up, the main episode.
Alright, folks, quick pause.
Here's the uncomfortable truth.
If your enterprise AI strategy is we bought some tools, you don't actually have a strategy.
KPMG took the harder route and became their own client zero.
They embedded AI and agents across the enterprise, how work gets done, how teams collaborate, how decisions move, not as a tech initiative, but as a total operating model shift.
And here's the real unlock.
That shift raised the ceiling on what people could do.
Humans stayed firmly at the center while AI reduced friction, surfaced insight, and accelerated momentum.
The outcome was a more capable, more empowered workforce.
If you want to understand what that actually looks like in the real world, go to www.kpmg.us slash AI.
That's www.kpmg.us slash AI.
This episode is brought to you by Mercury, radically different banking now available for personal accounts.
I already use Mercury for my business, so when they introduced personal accounts, it made immediate sense for me.
I try to bring the same level of intention to my personal finances that I bring to building companies, and most traditional banks just do not feel designed for that.
With Mercury Personal, you can toggle between business and personal in a click.
You can set up sub-accounts for specific goals, automate transfers so projects and savings fund themselves, and put idle cash to work with high yield savings, all without friction.
It's built for people who care about how their money moves and want tools that actually keep up.
Visit Mercury.com slash personal to learn more.
Mercury is a fintech company, not an FDIC insured bank.
Banking services provided through Choice Financial Group and Column NA members FDIC.
Today's episode is brought to you by Granola.
Granola is the AI notepad for people in back to 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 slash AI daily and use code AIDAily, new users get 100% off for the first three months.
Again, that's granola.ai slash AI daily.
If you're looking to adopt an agentic SDLC, Blitzy is the key to unlocking unmatched engineering velocity.
Blitzy's differentiation starts with infinite code context.
Thousands of specialized agents ingest millions of lines of your code in a single pass, mapping every dependency.
With a complete contextual understanding of your code base, enterprises leverage Blitzy at the beginning of every sprint to deliver over 80% of the work autonomously.
Enterprise grade end-to-end tested code that leverages your existing services, components, and standards.
This isn't AI Autocomplete.
This is spec and test-driven development at the speed of compute.
Schedule a technical deep dive with our AI experts at Blitzy.com.
That's B L I T Z Y.com.
Welcome back to the AI Daily Brief.
Vive coding is one of those terms that was extremely useful in drawing a hard line in differentiation from what came before, but which might ultimately have a fairly short shelf life.
It's been just 14 months since the term was coined by Andre Carpathi, and already many people have moved on from it entirely.
One of the reasons it doesn't quite fit anymore is that the difference between vibe coding and I don't know what you want to call it, artisanal by hand coding, have been increasingly obviated by the way that the coding platforms have evolved.
In other words, there is increasingly nothing but vibe coding.
Now, as we will see, mostly a process of multi-agent orchestration.
The other even more fundamental reason though that the term is running into some challenge is less about the vibe and more about the coding.
Does it make sense to even identify the activity as coding when coding is just the way that everyone builds everything?
Now I'm starting this conversation in the area of linguistics, but where it's showing up in practice is an extreme convergence in the actual product suites through which we interface with AI.
A few days ago, we got the latest theoretical leak out of Anthropic, and it looked like it might be some sort of full-stack vibe coding app along the lines of lovable.
In addition to having a very simple starting interface, it also looked like it came with a set of pre-built instructions for common tasks like scanning for security risks, exploring design directions, or implementing dark mode.
Now no one has confirmed this is real, but it didn't particularly surprise anyone, and Marmaduke 91 can on Twitter summed up most people's feelings when they wrote, Anthropic are coming after everything.
Now an Anthony Morris from Anthropic teased that a new Claude Code Desktop version would be coming out this week.
Vivecode's Riley Brown wrote, OpenAI and Anthropic are racing to get their version of Lovable out.
Anthony responded, My tweet has nothing to do with lovable though, laugh crying emoji.
And it turns out what Anthony was talking about was a more general Claude Code desktop update, which is what they released yesterday.
Anthropic's Felix Riesberg wrote, Today is a big day, we're launching a new version of Claude Code in the desktop app.
It's been redesigned for the ground up for parallel work and is a lot faster.
Now, in their announcement post, Anthropic emphasized that the redesign was focused on meeting developers where they are, which, as they acknowledge, is a completely different place from where they were six months ago.
They write, for many developers, the shape of agentic work has changed.
You're not typing one prompt and waiting.
You're kicking off a refactor in one repo, a bug fix in another, and a test writing pass in a third, checking on each as results come in, steering when something drifts, and reviewing diffs before you ship.
The new app is built for how agentic coding actually feels now.
Many things in flight and you in the orchestrator seat.
And indeed, in a lot of ways, the concept that runs through this is shifting the interface for a world in which development is orchestrating multiple agents.
This takes steps to make the Claude Code section of the Claude Desktop app into an agent orchestration command center.
Some of the features they point to include a new sidebar for managing multiple sessions at the same time, an integrated terminal and file editor, plus the ability to drag and drop to customize and arrange your workspace.
It certainly seems like one of the big focuses is the ability to run multiple sessions at the same time.
They write the new sidebar puts every active and recent session in one place.
You can kick off work across multiple repos and move between them as results arrive.
You can filter by status, project, or environment, or group the sidebar by project to find and resume sessions faster.
The first reports from the Anthropic team suggest that they've been shifting behavior to this new interface.
Cloud Code co-creator Kat Wu writes, Over the past few weeks I become a daily user of Cloud Code on desktop.
It's the best way to kick off and manage multiple CC sessions across local and cloud in one place.
Developer Daniel San, who had early access, pointed to the ability to run agents simultaneously as one of the things that he likes most about the new version.
He writes, the interface is much more tailored to a dev-oriented design, smaller inputs and more focus on agent execution and development tools.
Goyao pointed out, I think it'll be great to have multiple sessions be able to share context for the same feature.
I saw Cursor already start to do that, and I'm excited for all the labs to figure this out.
And indeed, this is certainly one of the big takeaways.
You might remember that at the beginning of the month, Cursor 3 gave developers the option to switch to a similar interface.
And for some, the convergence is clear.
Riley Brown again writes, Cursor Codex and Claude Code Desktop app look exactly the same now.
Now we'll have to see if that persists, because when Anthony Morris from Anthropic sent that tweet about the Cloud Code new desktop version, Andrew Abracino, who is a member of technical staff working on Codecs at OpenAI, replied, same.
Meaning I think that we can probably expect some at least small updates to Codecs coming this week.
Now, not everyone's first impression of the new Claude Desktop app was perfect.
Theo wrote that it froze on his first prompt, and I will admit it also completely imploded on a project that I was doing, where I had to go basically delete all the work that it had done and start fresh.
Still the biggest complaint, and something that you increasingly see around every post from Anthropic, continues to be about usage limits and throttling.
Control Alt Duane wrote, run multiple Claude sessions in this economy with these highly restricted usage limits?
It's nice in theory, but people can barely run one session with Opus 4.6 without chewing through usage, and they're saying you can run multiple?
Did they increase usage limits?
Gram VIP responded to the Claude announcement post and wrote, Allow me to fix your post.
You can now run multiple Claude sessions side by side for around five minutes before we lock you out and demand another 200 USD off you.
I think unfortunately for Anthropic, this is the big new PR battle, at least internally, that they're going to have to deal with.
Now there was one other Claude Code update yesterday as well.
Anthropics know as we been tweets, Claude code routines are here.
In addition to a schedule, you can now trigger templated agents via GitHub event or API with our infrastructure and your MCP and repos.
They've changed how we do docs, backlog maintenance, and more internally at Anthropic.
So routines basically are the next extension of the scheduled tasks feature which came a few weeks ago.
Rather than triggering at a certain time or day, routines allow tasks to be triggered via GitHub event or API.
In their docs, Anthropic wrote, a routine is a saved claude code configuration, a prompt, one or more repositories, and a set of connectors, packaged once and run automatically.
Routines execute on Anthropic managed cloud infrastructure so they keep working when your laptop is closed.
The register described routines as a dynamic cron job or a trigger-driven short-lived event.
Now, when you sit back and think about this, this seems like a big unlock for a number of different use cases.
Rather than requiring you to keep Claude running at all times, you can now offload complex and dynamic tasks onto Anthropic's architecture.
Now, on the one hand, each of these individual updates is small on its own, but in net you can see the path where they're going, where Claude Code is basically building out any possible variation of triggering an event to complete a task, with this addressing one of the big remaining limitations.
Greg Eisenberg from the Startup Ideas pod, as he is wont to do, immediately thought about how this could turn into new startup opportunities.
He writes, the model is the commodity, the trigger is the product, and whoever maps the most valuable real-world events to the most specific industry workflows is going to build something massive.
Here's what I mean by trigger.
A permit gets filed, a customer's usage drops 40% in a week, a competitor launches a feature, a deal sits in your pipeline untouched for 14 days.
Some public, some inside your own tools, and every single one is a moment where an AI agent can step in and do something valuable before a human gets around to it.
The playbook is like this.
Map every trigger that matters in one industry, wire an AI agent to do each one, sell the outcome.
The person who shows up first with exactly what someone needs at exactly the right moment wins the deal every time.
And the people who go embarrassingly deep on one industry's trigger map are going to build generational companies.
DeGregg it is one more indication of why this is, in his words, the greatest time to be starting a company.
Now, while we didn't get Anthropic's full-throated lovable vibe coding experience, we did get a bunch of updates from Lovable themselves.
First of all, the company has launched a new desktop app, which allows users to access local MCPs, keep track of multiple projects, and use native keyboard shortcuts.
Still, maybe the bigger update from Lovable came a couple of days ago and was the native introduction of payments.
Users can describe what they want to sell and how they want the payments function to work in natural language, and the AI agent will do the rest.
Linear COO Christina Cordova writes: You know where you're not vibe coding?
PCI level one.
Partnerships with several acquirers per country all over the world, global tax compliance, and the list goes on.
The argument is basically that this is one of those key areas that just needs to be built in for people to unlock the full value of vibe coding platforms for business.
Indeed, as lovable CEO Antoine Ossika writes, it's never been easier to go from product to an actual business.
We saw vibe-related updates in other parts of the market as well.
A small one from Google, the company launched a new design preview in AI Studios Vivecoding experience.
Logan Kilpatrick writes, now while you wait for an app to be built, Gemini will create custom themes you can easily choose from in seconds.
To me, it looks like a bit of an integration with what you can do in their dedicated Stitch app, but brought into the Google AI Studio experience, which is kind of the pattern that we've been seeing with Google, where they're both developing key products and features in their own experiences, but then also working to integrate them more seamlessly.
Google also announced a new feature called Skills in Chrome.
It allows users to save and reuse AI prompts when using the Gemini assistant in the Chrome browser.
Users can choose from a skills library or build their own skills with Gemini's help.
And while the name is the same, the feature is closer to a prompt library than the full skills file system implemented by Anthropic and OpenAI.
Still, it is spiritually in the same domain if focused on a slightly different audience.
Google says the feature allows users to build one-click workflows for frequent tasks and gave the example of calculating nutrition stats for a recipe, comparison shopping, or summarizing documents.
Now, interestingly, when it comes to some of these bigger themes of how the way that people interact with AI is changing, this came up in a recent interview with Google CEO Sindar Pchai, and it sounds like they're thinking about the implications not just for their AI tools, but for all of their core experiences.
In that interview, Pichai said, if I fast forward, a lot of what are just information seeking queries will be agentic search.
You'll be completing tasks, you'll have many threads running.
When asked then if traditional search will still exist in 10 years, he said, it keeps evolving.
Search would be an agent manager.
Over in Microsoft Land, we continue to hear about how the company is looking to claw if I pilot.
The information reports that Microsoft is testing features inspired by OpenClaw, with a statement from corporate VP Omar Shaheen saying that a newly created team is, quote, exploring the potential of technologies like OpenClaw in an enterprise context.
A number of folks picked this up a couple of weeks ago when Omar announced his new role, with Swix from Latin Space writing, wow, Microsoft just appointed a corporate VP whose entire job is OpenClaw.
Now, of course, one of the big challenges for OpenClaw is the issues surrounding enterprise usage of it.
As you've heard, we've been running the Enterprise Claw program, and part of that program is actually giving people general alternatives and helping them understand the security risks, but there are some big barriers there.
Microsoft clearly sees an opportunity in this challenge, with sources suggesting that Microsoft are convinced that they can implement a safer version of the product by limiting permissions and siloing work within certain roles.
I would actually anticipate the general vibe code hardening for the enterprise to be one of the biggest building opportunities of the rest of 2026.
And clearly I am not alone with a new release from the company Superblocks getting a lot of chatter on X.
Superblock CEO Brad Menizes argued that vibe coding apps have quickly become an attack vector in the enterprise.
Like shadow AI use in previous years, many companies are now seeing employees building apps on production data with very little or zero oversight from their IT team.
Menizes noted that with Mythos approaching, cyber threats are also about to get a huge boost.
He then frames their company's new Superblocks 2.0 platform as the quote, only platform to take back control.
And basically what this is is a way to make employee vibe coding safe.
Brad writes, business teams build AI powered apps with permissions baked in.
IT and security can audit everything and lock down anything instantly.
Engineering sets the standards, every app follows them.
Now it seems to me that most of the replies to this have the feel of undisclosed paid posts, but holding aside the specific example, it is very clear to me that enterprise hardening of vibe coding is going to be a huge trend for this year.
Take all this together, and while there's no one massive change that happened, there is a very clear trajectory here.
Not only are all the labs, both the agent labs and the model labs, racing to make all the improvements that make their agentic coding apps most useful 24-7, whether you're sitting there in front of them or not, you're also increasingly just seeing companies deal with the implications where all of knowledge work runs in some way through code.
The products are converging because coding to build things that do your work or help you do your work is becoming one of the core primitives of knowledge work.
It is super interesting to see how these things evolve.
For now, you got a bunch of new toys to play with, and I'm excited to hear what you build.
But that's gonna do it for today's AI Daily Brief.
Appreciate you listening or watching as always, and until next time, peace.