The BeAIReady Brief | Week 27
June 29 – July 5 | The Companies That Built the AI Giants Are Pricing Their Exit, Most Bosses Who Fired People for AI Now Regret It, and Amazon Turned Your Adoption Problem Into a Billion-Dollar Produ
Satya Nadella spent last week warning the AI industry about itself. One of the people most responsible for the current buildout — Microsoft is reportedly committing something near $190 billion to data centers and capacity this year — stood up and told the frontier labs that their promises of mass job loss and concentrated power weren’t going to hold.
He wasn’t the only one hedging: Amazon spent the week hunting for cheaper models, companies that fired people for AI spent it rehiring them, and the June jobs report landed Thursday soft — 57,000 new jobs against expectations near 110,000 — with an unemployment rate that ticked down to 4.2% mostly because people left the labor force, not because they found work. What tied last week’s reading together, for me, was the sound of the AI story’s loudest backers lowering their voices — and the widening distance between what AI was sold as and what organizations can actually do with it.
This week’s coverage:
The Giants Are Getting Too Expensive for Their Own Backers
The two companies that funded the frontier labs spent last week looking for the exit — one from a conference stage, one from a spreadsheet.
Microsoft Cut Deeper While Others Started Rehiring
Microsoft trimmed thousands to pay its AI bill, right as the data showed most companies that fired people for AI wish they hadn’t.
Amazon Will Sell You the Thing You Can’t Do Yourself
The distance between personal AI and organizational AI got wide enough for Amazon to build a billion-dollar business inside it.
A Faster Copilot Doesn’t Fix What’s Upstream
Microsoft made Copilot quicker and cleaner, which is real — and beside the point last week’s reading kept making.
Here’s what I was reading.
The Giants Are Getting Too Expensive for Their Own Backers
There is something almost too on-the-nose about Satya Nadella spending last week as the voice of AI restraint. This is the man whose company bankrolled OpenAI and is now reportedly pointing close to $190 billion at data centers and capacity in a single year. But the argument he made was sharper than the usual executive hedging. He told the frontier labs, in effect, that the story they’ve been selling — mass automation, concentrated intelligence, a handful of models running the economy — is neither sustainable nor politically survivable, and that the future belongs to a “frontier ecosystem” where companies use their own data, keep their own economic agency, and run cheaper, controllable models instead of renting power from a few giants. When the person who helped build the giants starts arguing that nobody should depend on giants, it’s worth asking what he see from where he sits. My read is that this is less a change of heart than a change of position — Microsoft does better selling controllable enterprise AI than it does subsidizing a lab that competes with it. Either way, my diagnosis is that Nadella is right — even if his motive is commercial. (Microsoft’s Nadella turns on the AI giants he helped build)
The same pressure showed up in Amazon’s ledger. Anthropic is renegotiating its arrangement with Amazon and moving to per-token pricing next year, a shift that could substantially raise what Amazon pays to run Claude-based services — and Amazon responded by shopping for cheaper alternatives, including, pointedly, OpenAI. This is the tell: the hyperscalers that made the frontier labs possible are starting to treat them as a cost line to be managed, not a partnership to be protected. Amazon has money in both Anthropic and OpenAI, which means it can afford to play them against each other, and the government scrutiny that has landed on Anthropic’s models lately only strengthens the case for spreading the risk. For enterprise leaders the lesson isn’t about Amazon’s balance sheet — it’s that the price of frontier AI is not fixed, single-vendor dependence is a real exposure, and building for model portability before your provider reprices you is starting to look less like paranoia and more like planning. (Amazon seeks cheaper AI alternatives as Anthropic shifts to token-based pricing)
Two years ago the frontier labs looked like the center of gravity. Last week their two biggest backers spent their time explaining, in different dialects, why they’d rather not be quite so dependent on them.
Microsoft Cut Deeper While Others Started Rehiring
Microsoft moving to cut thousands of jobs — under 2.5% of a roughly 220,000-person workforce, concentrated in sales, consulting, and Xbox — would be an ordinary restructuring story if it weren’t landing while the company’s AI and cloud spending sailed past $100 billion. The framing writes itself: the buildout is being funded, in part, by the people being shown the door. As a Microsoft partner I sit close enough to this to say the cuts aren’t simply “AI replaced these roles” — a lot of it is Xbox absorbing years of overspending and a sales org being reshaped — but the optics of trimming headcount to feed capex are hard to escape, and the unions now demanding to bargain clearly aren’t buying the nuance. The uncomfortable question underneath the layoffs isn’t whether Microsoft can afford the AI bill; it’s what return the $100 billion is supposed to produce, and by when. (Microsoft to Cut Thousands of Jobs Across Sales and Xbox)
Set that against the CNBC piece that highlighted a survey of business leaders, where 39% said they had cut jobs because of AI — and 55% of those admitted the decision was wrong. Ford, Commonwealth Bank of Australia, and IBM have all been rehiring humans to clean up what automation broke: quality problems, customer-service failures, the judgment calls software doesn’t make. The AI-first layoff, it turns out, was often a bet placed before anyone checked whether the technology could actually hold the job. I’ve been saying some version of this for a year now — that AI is spectacular in the hands of an individual and treacherous as a wholesale replacement for one. The companies rehiring aren’t conceding that AI doesn’t work. They’re conceding that they mistook a capable assistant for a seasoned employee. (Employers who laid off workers citing AI are already starting to regret it)
One company is cutting to place a bet; a lot of others are paying to unwind theirs. Both are the same story told from opposite ends — nobody is sure yet what this technology is actually worth to them.
Amazon Will Sell You the Thing You Can’t Do Yourself
If you want a single data point that proves the hardest part of enterprise AI isn’t the model, look at what Amazon just decided to spend money on. AWS is putting roughly a billion dollars into forward-deployed engineers — human teams that embed inside a customer, stand up agentic AI systems, transfer the skills, and then leave once the organization can run on its own. Early customers include the Allen Institute, Cox Automotive, the NBA, the NFL, Ricoh, and Southwest Airlines. Amazon has correctly identified that the bottleneck in AI adoption is organizational, not technical — and has turned that bottleneck into a billion-dollar services line. That is the whole argument of this newsletter, priced and productized: the model is available to everyone, and the thing that’s scarce is the capacity to redesign how work happens around it. The catch, which the piece names, is that this kind of high-touch help is built for companies that can afford a billion-dollar vendor’s attention — which does very little for the mid-market organizations where most people actually work. (Amazon is spending billions deploying engineers into customers looking to get started with AI)
The HR research out last week described the same problem from the inside. A survey of more than 1,300 business and HR leaders found AI in HR still amounts to “a series of disconnected experiments” — pilots at the edges, not workflows rebuilt around the technology. The organizations that did the harder work of building an actual AI culture reported up to 4.5 times the impact. Experimenting at the margins feels like progress and produces almost none of it, because the returns don’t come from adding AI to a process — they come from redesigning the process around it. The uncomfortable part for HR specifically is that this turns it into a design function rather than a support function, and most HR teams are neither staffed nor mandated for that. (HR is still ‘experimenting at the margins’ on AI, report says)
And where the organization doesn’t provide the tools, the training, or the rules, people don’t wait. Nearly a quarter of U.S. workers — 23% — are now using their own AI tools every day, sourced on their own and governed by no one. “Bring your own AI” isn’t really a security failure; it’s a symptom — employees filling a vacuum their employers left open, with every private tool a data-handling decision nobody signed off on. The reflex is to write a policy banning it, but that misreads the situation. The demand is already there. The only real choice is whether the organization meets it with sanctioned tools and training or keeps pretending the shadow economy of AI isn’t already running the place. (Rising ‘bring your own AI’ trend can spell trouble for employers)
Put those three together and the shape is unmistakable — the technology arrived, the organizational work to use it did not, and into that vacuum stepped a billion-dollar vendor, a stalled HR function, and a workforce improvising with tools nobody approved.
A Faster Copilot Doesn’t Fix What’s Upstream
Which brings me, finally, to Microsoft’s Copilot redesign — and to why I found it both useful and slightly beside the point. The new design is faster (load times cut by more than half, complex responses about 10% quicker), cleaner, and more task-aware, with a bigger prompt workspace and a single entry point that surfaces the right actions. Microsoft says usage across Word, Excel, PowerPoint, and Outlook climbed 27 to 43% after the change. As a Microsoft partner, this is exactly the kind of improvement I want to see — friction at the individual level is real, and reducing it matters, so I’ll take faster and clearer every time. But a better Copilot answers the one problem last week’s reading suggests isn’t the binding constraint: getting a single person from intent to output. The redesign optimizes the layer where AI already works. The layoffs, the rehires, the billion-dollar services bet, the shadow tools — those all live one level up, in the organizational layer a UX team can’t reach. A slicker prompt box is genuinely welcome. It just isn’t the thing standing between most companies and the returns they were promised. (Introducing a new design for Microsoft 365 Copilot)
The loudest people in AI spent last week arguing for less of it — less dependence, less automation, less certainty about what it replaces. Nadella wants a frontier ecosystem instead of a few giants. Amazon wants a cheaper invoice. The companies that fired people for AI want them back. And Amazon, reading the same room, decided the real money is in helping organizations do the work the technology can’t do for them. Strip away the specifics and last week said one thing plainly, at least to me: the constraint on enterprise AI was never the intelligence... it was the organization wrapped around it. That’s not a model problem or a licensing problem or a problem a faster Copilot solves. It’s a leadership problem — the unglamorous work of redesigning how people work — and it’s the one part of all this that no vendor, however large, can be paid to care about as much as you do.
That’s it for this week’s BeAIReady brief!
If you appreciate the depth of reporting and how I connect the dots, please like, share this post, and subscribe (or share the Brief with a friend!). Thanks!
~erick