Procurement's playbook just broke
Your AI vendor is now your consultant. Here is what to ask before the next vendor meeting.
Hey,
Here’s your AI and tech minute.
Safe to skip this week
Here’s what you can ignore, and the one thing that actually matters.
Skip: Cinematic Video Overviews from NotebookLM. The demos look great. The feature is locked behind the $250/month Ultra tier, capped at 20 a day, and most of your team will never touch it.
Watch: OpenAI and Anthropic both stood up consulting arms last week. Your vendor and your implementation partner are now the same company.
Your AI vendor is now your consultant
There used to be a useful line between the company that built your AI and the company that helped you install it. That line just got blurry.
Last week, Anthropic and OpenAI both launched consulting-backed implementation arms. Different PE firms. Same strategy: own the layer between the model and the enterprise. OpenAI’s new entity is called The Deployment Company. It raised more than $4 billion from TPG, Brookfield, Advent, and Bain Capital, with 19 investors total. Anthropic’s $1.5 billion firm is being built with Blackstone, Hellman & Friedman, and Goldman Sachs as founding partners.
Neither firm is building an advisory shop that sits alongside the model. Both are embedding their own engineering teams directly inside the new entity. OpenAI even has a name for these people: Forward Deployed Engineers. They sit inside the customer’s office, scope the work, and build the systems. The implementation team is, in a literal payroll sense, the model team. That changes the incentive structure completely.
What changes for procurement
The vendor whose tool you are evaluating will also be the firm that scopes your deployment, builds the integrations, and consults on the governance review. The procurement question is no longer “is this the right model?” It is “Who is checking the work?”
Three shifts worth tracking:
Independence is weaker. The engineer building your workflow has incentives to recommend their own model, agents, and tooling. Not a scandal. Economics.
The “implementation partner” check is degrading. Companies that relied on a separate systems integrator as a sanity check may find that integrator owned, acquired, or outbid by the model provider.
Switching costs are about to rise. The deeper the deployment team embeds, the harder the next model evaluation gets.
Stuck-in-pilot midmarket companies finally get engineers on-site. That part is real. But the procurement playbook needs an update.
If you only have 90 seconds with procurement, ask three questions: Does the team scoping this work have an internal incentive tied to model selection? What is the exit cost if you switch models in 18 months? Who is the second set of eyes on the architecture? If the answer is “us,” that is not a second set of eyes.
Tool of the Week: NotebookLM (the half we didn’t cover the first time)
We featured NotebookLM back in Issue #5 in January, but only the audio overviews piece. The part that turns a document into a two-host AI podcast. It was a fun feature. It is still a fun feature. It is not what our team uses it for now.
Since January, Google has shipped a stack of upgrades that turned NotebookLM into something else entirely. The audio thing turned out to be the gateway. The actual workhorse is everything else.
What we use now:
Mind Maps that visualize how your source material connects
Data Tables that pull scattered information across documents into a structured comparison table, exportable to Sheets
Reports that draft briefing docs, summaries, and study guides from your stack
Slide decks built from your sources with PPTX export and per-slide revisions
Saved chat history that finally makes it usable as a real working notebook
The core promise is unchanged. Drop in a stack of sources (PDFs, links, docs, transcripts, EPUBs, images, and CSVs), and the model reasons against only those documents. Not the open web. Not its training data. Your stack.
For corporate readers, that is the difference between “what does the internet think about our Q2 strategy?” and “what does our actual Q2 strategy doc, board pre-read, and customer interview transcript say about Q2?” The first is theater. The second is analysis.
Our team uses it for vendor diligence packets, contract redlines, multi-doc research, and prepping for a meeting where you need to know the source materials cold.
Free. notebooklm.google.com
Hype Check: “Frontier companies use 3.5x more AI per employee”
That stat is from OpenAI’s new B2B Signals report, which aggregates anonymized usage data from OpenAI’s own enterprise customers. It is being passed around as proof that AI delivers productivity at scale.
A few things to notice. “Frontier” is OpenAI’s term. The comparison group is “typical firms” using OpenAI products, not firms using any AI at all. The report measures token consumption, not business outcomes.
Verdict: Overhyped. Token consumption is an input, not a result.
If you want a real measure, track these instead: number of workflows automated, time-to-resolution on internal tickets, or revenue tied to AI-assisted deals.
One-Minute Win: The Dashboard Check
Most large employers now track which employees use AI, how much, and on what. Most employees don’t know what shows up on those dashboards. Sixty seconds to find out:
Open your IT or admin policy doc. Search “AI,” “monitoring,” “Copilot,” and “ChatGPT.”
Ask IT or your manager what usage data leadership sees and on what cadence. Phrase it as “I want to make sure I’m using the tools right.” That framing is safe and gets you the same information.
Compare your usage pattern to what the policy says is “expected” or “encouraged.”
You are not trying to game the dashboard. You are trying to know whether one exists, what is on it, and whether your name is on the right side of it.
Quick Hits
Workplace: The 2026 WRITER enterprise AI survey landed last week. 54% of C-suite executives say AI adoption is tearing their company apart. 75% admit their AI strategy is “more for show.” 48% now call AI adoption a massive disappointment.
Workplace: CNBC reported last week that most large employers are now tracking which employees use AI, how much, and on what. The dashboards are quietly going live.
Policy: Microsoft, Google, and xAI agreed to give the Commerce Department’s Center for AI Standards and Innovation early access to frontier models for pre-deployment review. CAISI has already completed more than 40 frontier-model evaluations. Another answer, in a different way, to “who is checking the work?”
Enterprise: Microsoft shipped Agent 365, a control plane and governance layer for AI agents across Microsoft 365. For IT teams sitting on agent rollout, this is the procurement signal they have been waiting for.
Policy: On paper, the Colorado AI Act takes effect in 49 days (June 30). In practice, a federal court paused enforcement on April 27 pending litigation, and lawmakers are negotiating to push the date to January 2027. The law is real. The enforcement window is, for now, closed. Most law firms are still telling clients to prepare for June 30 anyway.
That’s your minute.
See you next Tuesday.
The AI Minute
getaiminute.com
Know someone drowning in AI and tech noise? Forward this to them.
We read 47 newsletters so they don’t have to.