My AI Budget Just Shrank by 98%

Hello,

This is Simon with the latest edition of The Weekly. In these updates, I share key AI related stories from this week's news, list upcoming events, and share any longer form articles posted on the website.

We are a very AI-forward company at work, which is understandable given that we are a software company with AI at the core of our product. We have been encouraged to use it from day one, and the company has provided clear guidance around the terms of use at work. The business recently decided to pick Anthropic and their Claude solution as our enterprise AI tool of choice. Originally, we noticed that our personal usage limit appeared to be "unlimited". This became $15,000, and then last week it was $1,000, and just today it went down to $300 per month as a basic default. This new limit has already caused a lot of problems for many people, and they have asked for it to be increased again.

I have a couple of observations on this. Firstly, when the limit was unlimited, we didn't really pay any attention to what we were spending — it didn't really seem to matter. It wasn't that everyone was trying to use as many tokens as possible, but it seemed arbitrary. As soon as an actual cap was put in place, it instantly made people think about what they were spending, as they now had a percentage metric that showed you've used "16% of your monthly allowance". This reinforced the idea that it's not an endless resource, and its use should be thought about.

The second observation is that $300 is actually pretty low for most people. I saw dozens of people instantly put their hand up asking for it to be increased, as it was "vital for their role and the work they were doing". Which I can believe it is.

But all of this leads back to the question: what is the return on investment (ROI) on our AI spend? I won't tell you how much I've been spending, but it's somewhat modest in comparison to some of the initial limits that were set. Now that I can actually see how much running a specific task costs, though, you really do start to wonder if it's worth it.

There are more and more articles talking about this very issue. Now that AI usage is ubiquitous across every organisation, CFOs across the globe are staring at ever-larger invoices, and questions are being asked.

There's some research by Forbes that suggests less than 1% of executives report ROI of 20% or greater, and only just over half, 53%, report 1-5% ROI. Those are not impressive numbers. We're clearly all in an experimentation phase right now, with seemingly very few companies generating any sort of decent return on their AI spend. Whilst it might be possible to write this off at the moment, CFOs, shareholders and investors only have a certain amount of patience.

Do you have any idea whether your AI use is worth it? Does your company track what you're spending and compare that to the outputs?

Your documentation is a product decision. The companies building the most-used developer tools in the world chose Mintlify because great docs reduce support load, accelerate adoption, and convert evaluators into customers. Backed by a16z and Salesforce Ventures, Mintlify powers 20,000+ companies with AI-native documentation that keeps pace with your product — without pulling engineers off the roadmap to maintain it.

Book a demo

Exclusive for subscribers.

In this section, I’m going to bring to you a real world example of AI use.

Microsoft has launched a new operating business called Microsoft Frontier Company, committing $2.5 billion and 6,000 industry and engineering experts to embed directly inside enterprise customers and deliver AI deployments that produce measurable outcomes. Announced on 2 July, the unit will co-design and run AI systems alongside clients, drawing on industry specialists, change managers, and technical consultants. Early partners include the London Stock Exchange Group, Unilever, and Accenture. The move came two days after Amazon announced a similar $1 billion commitment, and joins comparable ventures from OpenAI and Anthropic — meaning every major AI vendor now has a professional services arm focused on making deployments actually stick, according to CNBC and GeekWire.

Why it matters: The rush to embed engineers inside customers is a tacit admission that selling AI software is no longer enough — the real problem is getting it to work at scale. For organisations that have invested in AI tools but are struggling to see returns, it signals that the vendors are now in the implementation business too, which has significant implications for how procurement, IT teams, and transformation programmes are structured.

The EU Council gave final approval on 29 June to the Digital Omnibus package, which pushes back the compliance deadline for high-risk AI applications from August 2026 to December 2027. Systems covered by Annex I, which is those touching critical infrastructure and safety components, get even more time, until August 2028. The transparency obligations from the original AI Act still take effect in August 2026 as planned. The change follows lobbying by businesses who argued the original timeline was unworkable, according to DLA Piper and Gibson Dunn briefings.

Why it matters: Any organisation that has been building AI compliance programmes around an August 2026 deadline now has some breathing room, but the risk is that teams use the extension as an excuse to delay. Transparency rules still kick in this summer, and the EU Act is increasingly being used as a template by regulators elsewhere, so the direction of travel has not changed, only the pace.

PwC's 2026 Global AI Jobs Barometer, analysing over a billion job advertisements across six continents, found that AI is creating a two-track labour market. "Professionalised" roles where AI handles routine tasks and human judgement is amplified, are growing twice as fast and seeing salary growth 42% faster than roles being "democratised" by AI, where the technology simply makes the work easier for anyone to do. Roles requiring AI skills commanded a 62% wage premium in 2025. Junior roles in the most AI-exposed sectors are now seven times more likely than their least-exposed equivalents to require leadership-level skills, according to PwC.

Why it matters: For managers and HR teams, the findings reframe the AI skills question and it is not about which jobs survive, but about whether your workforce is on the professionalising track or the commoditising one. The report suggests that organisations investing in AI are growing headcount and wages faster, while those that are not are falling behind on both.

Thanks for reading, and see you next Thursday.

Simon,

Was this email shared with you? If so subscribe here to get your own edition every Thursday.

Enjoying Plain AI? Share it and get a free gift!

If you find this newsletter useful, why not share it with a couple of friends or colleagues who would also benefit? As a special thank you, when two people subscribe using your unique referral link, I’ll send you my "Exclusive Guide: Supercharge Your Work with NotebookLM." It’s a practical, no-nonsense guide to help you turn information overload into your secret weapon.

You can also follow me on social media: