Plus: Google built a CLI for agents, Anthropic is suing the Pentagon, and OpenAI's agents can now use your computer.
Mindmaker LiveEdition #1 · March 10, 2026
Plug-and-play AI is
officially a myth.
That's not me saying it - it's 600 AI decision-makers surveyed by Cognizant, published today. 63% of enterprises report a "moderate-to-large gap" between their AI ambitions and what they can actually deliver. And these aren't companies ignoring AI; 52% are spending over $10 million a year on it.
The money is there. The capability isn't. That gap is the story of 2026 so far.
The Signal
4 things that moved this week
1
Citi just raised its AI capex forecast by nearly a trillion dollars
Citigroup bumped its 2026-2030 global AI capex estimate from $8T to $8.9T, and revenue forecasts from $2.8T to $3.3T. Enterprise adoption is moving faster than anyone modeled. Hyperscalers alone are on track to spend $630B this year. If you're wondering whether the AI investment wave is real or hype - it's real, and the institutional money just told you so.
2
Google shipped a CLI for all of Workspace - built for agents first
Google released gws, an open-source tool that lets AI agents manage Gmail, Drive, Calendar, Sheets, and Docs programmatically. Built in Rust. Designed for agents first, humans second. This is Google signaling that the future of productivity software isn't people clicking buttons - it's agents running workflows. If your ops team hasn't started thinking about "agent-ready infrastructure," this is your nudge.
3
Anthropic is suing the Pentagon
The Pentagon designated Anthropic a "supply chain risk" - the first time a US company has ever received that label. It followed Dario Amodei's refusal to allow Claude for autonomous weapons or mass surveillance. Trump ordered federal agencies to ditch Anthropic tech. If you're using Claude in any workflow adjacent to government contracts, this matters now. The company building one of the most capable models on the planet just became a geopolitical football.
4
OpenAI's GPT-5.4 can now use your computer
GPT-5.4 launched with native computer-use: it controls mouse and keyboard from screenshots, browses the web, works across apps. 75% success on desktop navigation (humans score 72.4%). 1M token context. The first general-purpose model that genuinely operates a computer like a person. Not perfectly - but well enough to replace multi-step manual workflows nobody wanted to do anyway.
Under the Hood
The plug-and-play myth
Here's what caught my eye in the Cognizant study, beyond the headline.
They surveyed 600 AI decision-makers and interviewed 38 executives. The finding that stopped me: IT services firms hold a 23% trust advantage over management consultancies when it comes to actually scaling AI into business operations.
That's a reversal. For the past three years, McKinsey and BCG have dominated the "AI transformation" conversation. They write the reports, run the workshops, build the roadmaps. But when it comes to making AI work in production - actually deploying, integrating, and maintaining systems - enterprises trust the people who build things over the people who advise on things.
84%
of enterprises now maintain formal AI budgets
91%
expect AI budgets to grow in the next two years
31%
still can't demonstrate ROI from AI investments
27%
cite talent shortages as their primary barrier
Everyone's spending, most can't prove it's working, and the gap between ambition and execution is the defining challenge.
This matches what I'm seeing in conversations with leaders running AI initiatives at mid-market and growth-stage companies. The problem isn't "should we use AI?" That question is settled. The problem is "how do we make the AI decision that actually moves the business forward instead of creating an expensive experiment?"
The companies closing the gap aren't the ones with the biggest budgets. They're the ones making fewer, sharper decisions about where AI creates real leverage - and then going deep on execution in those specific areas rather than spreading thin across every possible use case.
The Decision
What this means for your next move
On AI strategy
Stop trying to be an "AI company." Pick the 2-3 workflows where AI creates measurable leverage (revenue, speed, cost), build or buy for those specific cases, and prove ROI before expanding. The 63% with capability gaps are mostly companies trying to do everything at once.
On your next software purchase
Your business tools are being rebuilt for agents, not people. This doesn't mean you need to act today, but your next purchase should include: "Can an AI agent use this without a human in the loop?" If the answer is no, you're buying something that will be legacy in 18 months.
On vendor risk
If Anthropic can get labeled a supply chain risk overnight, your AI vendor strategy needs a backup plan. Not because Anthropic is going anywhere, but because the political risk of depending on a single foundation model provider just went from theoretical to concrete.
What's the AI decision sitting on your desk right now?
Hit reply and tell me. I read every response.
Mindmaker Live
AI decisions that matter. Weekly. Written by Krish Raja.
If someone forwarded this to you, subscribe here.
Too much email? Unsubscribe.

