top of page
Search

Stop Buying AI. Start Building Value.

We’re back to the Summer of Copilot – my ongoing series as part of my Copilot Navigator newsletter on LinkedIn. 



If you lead a team right now, the AI headlines are sending you two messages. One says the technology is moving faster than you can plan for. The other, quieter one, is about leverage, cost, and control, and it is the one worth your time. Build 2026 was covered as a parade of agents. Look past the parade.


AI News – June 2026

Copilot Cowork

Copilot Cowork hitting general availability should have been a milestone. Instead it landed in a predictable fury over pay-as-you-go pricing.


That reaction skips the obvious. As token economics shift and AI workloads get heavier, the cost of delivering good intelligence is going up, not down. Pricing below cost at scale isn't just unsustainable. It can drift into predatory territory under U.S. antitrust law.


The common complaint goes something like this: "I had AI spend an hour reviewing a year of reports, and it cost me $15." Sounds steep, until you check the open market. What would you actually pay a capable analyst or consultant to do that same work, at that speed, with that consistency, and no backlog?


So this isn't really a pricing problem. It's a value reset. We're finally being asked to reconcile what the work is worth instead of what we're used to paying for software.


Plus, with the flurry of Cowork metadata skills being shared on social media, I’m still surprised when Anthropic users are surprised to learn that at Copilot already includes Opus 4.8, Sonnet 4.6, and Cowork.


Build 2026

The headline was not agents, it was independence.


MAI Models

I will permit myself one victory lap: last October, in our annual predictions, I called the date of this announcement to the day. Microsoft AI launched a family of seven in-house models, the MAI family, built across the tasks that show up in enterprise work:

  • MAI-Thinking-1, the flagship reasoning model, which Microsoft says matches leading models in its weight class.

  • MAI-Code-1-Flash, an efficient agentic coding model wired into GitHub Copilot and VS Code.

  • MAI-Image-2.5, MAI-Transcribe-1.5 across 43 languages, and MAI-Voice-2, each with a lower-cost Flash variant.


The scorecards are interesting. The strategy underneath them is what should change your planning.


For two years, Microsoft’s AI story rested heavily on a single partner. With MAI, Microsoft has its own engines sitting alongside OpenAI’s and Anthropic’s. In one move, it repositioned itself from a reseller of someone else’s intelligence to a more neutral provider of several engines. For you, that is optionality: match the engine to the task and the budget, and switch without re-platforming.

  • Clean data lineage. Microsoft says the MAI models are trained on traceable, enterprise-grade data with no distillation from other labs.

  • Frontier Tuning. Tune an MAI model on your own workflow data, in your own environment, and keep the result. A model trained on your institutional knowledge, owned by you, at lower cost, is a new and serious option.


Scout and the age of Autopilots

The second announcement was Microsoft Scout, an always-on agent in a category Microsoft calls Autopilots: agents that work autonomously on your behalf, with their own identity, without being prompted each time. An assistant helps while you are present. A worker holds your priorities and follows through. Scout can read and write files, run scripts, patch code, launch sub-agents, and operate a browser.


That power is the point, and it is also the risk. Scout is built on OpenClaw, the open-agent framework I wrote about earlier this year as a virus with a friendly face in its unmanaged form. What makes Scout notable is how Microsoft wrapped that capability in governance – what’s been dubbed “ClawPilot”:


  • Each Scout instance runs under its own governed Entra identity, so its actions are attributable to a real entity in your directory.

  • Credentials are scoped to individual tasks and redacted from diagnostic logs.

  • Activity is bound by Microsoft Purview sensitivity labels and DLP policies, and sensitive operations require human sign-off.


This is the governed-versus-shadow-IT choice I keep putting in front of leaders, and Microsoft made it explicitly. If your people want this, and they will, give them the governed version before they find an ungoverned one.


Enterprise AI News

Follow the money, and the compute

June has been a big month for AI news, not just Copilot Cowork reaching general availability.  Mostly it’s been about SpaceX. Are they an ISA company? You decide:



That last point changes the clock. Equity is patient. Debt is not. When a provider borrows billions, it has to service that debt on a schedule, which means showing returns sooner. That pressure travels downhill as pricing pressure, and you are downhill. The control points consolidating around models, compute, and distribution are the ones I mapped in The Silicon Seven. This is also why Microsoft’s MAI move matters to you: a credible in-house engine family is leverage against exactly this kind of pricing pressure.


In other news, Anthropic released Fable 5 and Mythos 5, and was forced to withdraw them three days later by the Trump Administration.  Bottom line – never rely a single model.


TechCon 365 Chicago


Your real scorecard: access is won

At TechCon 365 in June, we ran our 2026 AI Adoption and Governance model live with 18 leaders. Access to chat tools was the highest-scoring dimension by a wide margin. Then the honest part: the average score was 257 out of 500, and not one organization reached the next tier.


Where the value lives: mind the 161-point gap

Access to chat scored high. Putting autonomous agents to work scored far lower, a 161-point gap. Seven of eighteen used no agentic tools at all. That gap is where your return is hiding, and it is exactly the gap Scout and the Autopilots are built to close. The chat box is the on ramp, not the destination.


Is it working, and what does it cost?

Eight of eighteen do not measure the ROI of AI at all, and only two measure revenue impact. The moment you can put time saved or revenue moved next to a workflow, you change the conversation with your CFO.


License spend is easy to track. Variable spend is not, and you are about to have more of it. Set a token budget per workflow, turn on alerts before the invoice, and name an owner.


Governance is your steering wheel, not your brake

Governance was the strongest dimension overall, until you see what is carrying it: a published acceptable-use policy. Thirteen of eighteen named IT and the help desk as their main AI support model; only two had a Center of Excellence. A help desk can reset a password. It cannot teach your finance team to rebuild the monthly close around an agent.


Synozur CEO Michelle Caldwell also led several workshops in Chicago about AI adoption programs.  She gathered some fascinating data about the rise and fall of different personality types for AI program success – and we’ll cover her research here as soon as its published.


Why pilots stall, and what the 5% do

Finally, MIT reports that most generative AI pilots never make it past the demo, and it is almost never the model. The teams that reach the P&L rank a use-case backlog from real work, put governance in place before they scale, and assign a named human owner to every AI output.


What to do this month

  • Kill nine of your twelve pilots. Keep the ones that directly drive revenue.

  • Measure something real. Pick two workflows and track time, cost, and outcome.

  • Put a number on variable spend before it puts a number on you.

  • Build the scaffolding. A Center of Excellence and real training, not just a help desk.

  • Use your new optionality. Run an engine bake-off across Microsoft, OpenAI, and Anthropic models, and let cost and quality decide.


The organizations that break into the next tier this year will not be the ones with the most licenses. They will be the ones who can answer three questions: What did it return? What did it cost? Who owns making it better? If you want to see where you land, our AI Maturity Model is open and free.


Thanks for joining us again.  Happy (Copilot) Summer.

bottom of page