You might notice this edition looks a little different.

We’re changing The AI Newsroom to focus less on AI news for the sake of keeping up, and more on developments that can help you make money, save money or save time.

That means fewer headlines and more judgment about what’s actually worth testing.

I hope you like the new format. And I mean that. If something isn’t useful, tell me.

This is the kind of AI case study I pay attention to.

OpenAI recently published a look at how German tax advisory network HSP GRUPPE is using ChatGPT across tax advisory, legal research, client communication, financial analysis and internal knowledge sharing.

HSP reports 84% weekly active usage, more than 500,000 ChatGPT conversations over roughly six months, and 98.6% of surveyed employees reporting higher productivity. The firm estimates AI has created about 40,000 hours of additional annual capacity across its network.

One partner said evaluating multiple real estate investments previously took around nine hours. With ChatGPT, she can prepare the analysis in about two.

Those are company-reported results, so I wouldn’t assume another business can replicate them. But the implementation is more interesting than the headline numbers.

HSP created shared AI workflows for repeatable work, including client communication and accounting questions. It also runs monthly forums where employees share successful use cases. Human professionals remain responsible for reviewing the work.

Here’s what I think matters: most companies are still treating AI as a collection of individual productivity tools. Someone discovers a good prompt, saves 30 minutes and moves on.

That’s useful, but it doesn’t scale.

I’d ask employees one question: What task have you already made substantially faster with AI?

Find the best answers. Turn them into repeatable workflows. Share them across the company.

That’s a more interesting AI strategy than buying another enterprise platform.

Zapier is moving its standalone Agents functionality into AI by Zapier inside regular Zaps.

The product detail matters less than the direction.

The winning “AI agent” for many businesses may simply become an intelligent step inside an ordinary workflow.

A lead arrives. AI researches the company and decides how to classify it. Regular automation updates the CRM. AI drafts the follow-up. A human approves it when necessary.

Some steps need judgment. Most don’t.

I’d ignore a lot of the hype around fully autonomous digital employees for now. For founders and operators, small amounts of AI judgment inserted into reliable processes may be much more useful.

Look for one workflow where your automation currently gets stuck because a human has to read, interpret or decide something. That’s where I’d test AI.

A recent study examined tens of thousands of Microsoft engineers during the company’s early-2026 rollout of command-line coding agents.

Researchers estimate adopters merged roughly 24% more pull requests than they otherwise would have.

That doesn’t prove they created 24% more business value. The researchers themselves note that merged pull requests are only a proxy for output.

But that limitation points to the useful lesson.

Stop measuring AI adoption. Measure output.

I’m less interested in how many employees use your AI tools than whether the tools are improving an outcome that matters.

For sales, did qualified meetings increase? For operations, did processing time fall? For developers, did useful work ship faster?

Pick an outcome that matters to the business. Then see whether AI moves it.

Otherwise, adoption can become a vanity metric.

Wispr Flow turns your voice into clean, professional text inside any app. Emails, Slack, client updates — speak once, send without editing. 4x faster than typing.

Try Wispr Flow free

I’m increasingly interested in the businesses being built around existing AI technology rather than the companies trying to invent the technology itself.

Reinvent Telecom offers a useful example. After evaluating more than 20 voice-AI vendors, it built a white-label AI receptionist on Telnyx that its reseller partners can sell under their own brands. Reinvent reported closing its first order on launch day.

I wouldn’t take that as proof that everyone should start selling AI receptionists.

The broader model is what’s interesting.

A dentist, HVAC company or property manager probably doesn’t want an “AI agent.” They want fewer missed calls, faster lead response or less repetitive admin.

That creates a potential service opportunity: take existing AI infrastructure, configure it around one expensive problem for one niche, and sell the outcome.

I’d rather test “we help HVAC companies respond to every missed call” than “we build AI agents.”

The technology is becoming easier to buy. Understanding the customer’s workflow is still valuable.

A while back, I asked whether you’d attend a free workshop on using AI to save five hours a week. A handful of you said yes.

Rather than turn it into a live workshop right now, I’m thinking about recording a practical video instead, showing workflows and tools I actually use to save time.

Building it would force me to separate what genuinely saves time from what merely looks clever.

If I make it, I’ll share it here for free.

What repetitive task would you want me to tackle? Hit reply and tell me.

If this gave you one useful idea, pass it along to someone who’d appreciate it.

Jason