Why OpenAI and Anthropic Are Becoming Consultants

May 5, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
May 05, 2026

HEADLINES

White House Reverses Course on AI Model Oversight

The Trump administration is weighing a formal government review process for new AI models before public release — a stark reversal from day one, when they revoked the Biden-era executive order that made safety testing mandatory and dismantled the US AI Safety Institute's authority. The shift is directly attributed to Mythos: the White House is reportedly concerned about political blowback if a devastating AI-enabled cyberattack occurs. Then, this morning, news broke that the Commerce Department has already reached agreements with Google, Microsoft, and xAI for early model access; notably, OpenAI and Anthropic have had similar agreements with Commerce going back to 2024. It's not yet clear whether these are the same as the cybersecurity-focused executive order being discussed, so there will likely be more to cover before the week is out.

The reporting triggered a wave of ourtage. Critics like Beff Jezos warned that future administrations will use model approval to bake in political bias. China Hawks like Chris McGuire at CFR called it a sorely needed pivot. Former Trump AI advisor Dean Ball and his Biden counterpart Ben Buchanan published a joint NYT op-ed arguing that model capability isn't actually the right regulatory target — the way models are used is the primary issue — and calling instead for mandated independent audits of developers' safety claims. Ethan Mollick flagged the deeper problem: our benchmarks for AI risk are so poor that it's hard to write vetting criteria that aren't hopelessly vague. To illustrate why, recent UK AI Safety Institute testing found that GPT-5.5 has roughly the same automated hacking capabilities as Mythos — which makes it very unclear where any government would even draw the line.

MAIN STORY

No AI Transformation Without Org Transformation

Both OpenAI and Anthropic announced major enterprise consulting ventures on the same day, each backed by billions from private equity. The timing isn't a coincidence — it's an admission that the last mile of AI deployment is proving far longer than anyone expected, and that selling models isn't enough. The real bottleneck isn't capability; it's that most organizations simply aren't structured to absorb AI. New data from Microsoft's annual Work Trend Index puts numbers to what practitioners have been observing: organizational factors — culture, manager support, talent practices — account for more than twice the impact on AI outcomes as individual mindset and behavior.

OpenAI's Deployment Company
A $10B venture to tackle the "last mile" of enterprise AI.
Bloomberg reported that OpenAI has raised over $4B from PE firms including TPG, Brookfield, Bain Capital, and Advent for a new venture called The Deployment Company, initially valued at $10B and majority-owned by OpenAI. There's been no formal announcement yet, but from what's circulating, this isn't a sideshow — it cuts to the core of OpenAI's new enterprise focus and reflects a clear-eyed recognition that getting AI to actually work inside organizations is a fundamentally different challenge from building the models.

Anthropic's Enterprise AI Services Company
$1.5B with Blackstone, Goldman, and Hellman & Friedman — FDEs as the model.
Anthropic's venture is more formally announced: Blackstone and Hellman & Friedman are each putting in $300M, Goldman Sachs $150M, with Apollo, GIC, and Sequoia among the other participants, totaling $1.5B. The model is forward-deployed engineers — Anthropic applied AI staff working alongside a customer's own engineering team to identify where Claude can have the most impact and build custom solutions. Anthropic is explicit that this isn't replacing their Claude Partner Network (which covers the big GSIs like Accenture, Deloitte, and PwC for large enterprises) but going deeper, with a focus on mid-sized companies. Their example is instructive: a healthcare network where clinicians spend hours daily on documentation and compliance — the engagement starts with engineers sitting alongside clinicians to understand where time actually disappears, then building around that knowledge. There's an embedded thesis here about how AI transformation diffuses through a company: not top-down via an AI team, but through new working partnerships between engineering and every other function.

Sierra
The original FDE model, now raising at $15B.
Brett Taylor's customer and client relationship agent company Sierra announced it's raising just under $1B at a $15B+ valuation. Sierra's operating model has always been support-heavy deployment engagements — sitting somewhere between a tech startup and a consulting firm — which makes it useful context for understanding what OpenAI and Anthropic are attempting to scale. Sierra is essentially the proof of concept that this kind of embedded, co-building engagement model actually works at commercial scale.

The AI Consulting Boom
The consensus reaction to both announcements was essentially: "obviously."
The AI consulting gold rush is already well underway — there are reportedly more FDE job openings right now than there are FDEs currently employed in the US. Aaron Levie at Box argues the implementation work alone will exceed anything we currently imagine, given just how many moving pieces are involved in taking a company from a chat paradigm to agents that participate in meaningful workflows. The deeper point, which Drew Bredvick made back in October, is that model and capability innovations are structurally likely to outpace the ability of organizations to absorb them — which is precisely the gap these ventures are trying to close.

The Microsoft Work Trend Index
"Most organizations are not yet built to capture the value of expanded human agency."
Microsoft's annual Work Trend Index provides the data layer behind this whole story. They divided their findings into three sections: AI's lift on individual potential, the imperative for leaders to re-architect work, and the case for every firm to become a learning system. The most striking framing is a quadrant they call the Transformation Paradox: only 4% of organizations have both high individual AI capability and high organizational readiness (what Microsoft calls the "frontier"). 16% are stalled on both dimensions. The biggest and most underreported category is what they call "blocked agency" — high individual capability, low organizational readiness — where motivated employees are effectively constrained by the organizations around them. The headline finding from section three is that organizational factors (culture, manager support, talent practices) account for more than 2x the impact on AI outcomes versus individual mindset and behavior. In the agent era, this gap is getting worse, not better: the number of active agents in Microsoft's ecosystem has grown 15x year over year, and 18x in large enterprises.

Microsoft Work Trend Index

The Three Adoption Shortcuts That Don't Work
Buy and Hope, Contain and Delegate, and outsourcing the knowledge.
AIDB training leader Nufar Gasbar identified three common patterns that organizations fall into but that reliably fail. The first is Buy and Hope: pay for the licenses, send the excited email, and hope transformation follows. The second is Contain and Delegate: hand the transformation mandate to the AI team rather than engaging with it yourself and diffusing it broadly. The third — which Ethan Mollick has beat the drum about — is outsourcing the knowledge: hiring a GSI or consulting firm to "figure it out" on your behalf. There's clearly a role for external expertise partnerships (both Anthropic and OpenAI are raising billions to prove it), but the model isn't dropping in an expert to solve it for you — it's embedding builders alongside your teams to run a best-practice process that integrates engineers with everyone else.