DeepMind CEO says he talks to Google CEO “every day” as the lab steps up competition with OpenAI

LONDON / MOUNTAIN VIEW — Google DeepMind CEO Demis Hassabis has intensified his day-to-day alignment with Alphabet CEO Sundar Pichai—saying the two speak “every day”—as Google accelerates its push to turn frontier AI research into product velocity and enterprise wins amid pressure from OpenAI and other fast-moving rivals.

The tighter cadence reflects a broader shift at Google: DeepMind is no longer a semi-detached “moonshot lab,” but the engine room for Gemini, agentic assistants, and next-generation AI experiences across Search, Android, Workspace, and Google Cloud.

The Brief: Where the technology is right now—and who’s adopting it

1) The tech has moved from “chatbots” to “agents that do”

The state-of-the-art is now defined by multimodal, tool-using models that can interpret text, images, audio/video, and increasingly take actions across apps and the web. Google has positioned this direction explicitly as a “universal AI assistant,” with Project Astra capabilities feeding into Gemini Live and other product surfaces.

In practice, this is the transition from:

• Q&A → workflow execution (book, buy, resolve, build)
• single prompt → multi-step planning + memory + real-time context
• one-off demos → integrated enterprise systems (CX, commerce, IT ops)

2) Adoption is widening, but it’s splitting into two tracks: productivity vs. “agentic operations”

Productivity layer (knowledge work): Microsoft has pushed Copilot hard through its enterprise footprint, including large-scale deployments via major services firms (Cognizant, Infosys, TCS, Wipro) totaling 200,000+ Copilot licenses—a marker that “AI in the flow of work” is becoming a default expectation.

Agentic operations layer (customer-facing + transactional): Google Cloud is now packaging Gemini Enterprise for Customer Experience as an agentic solution that blends shopping and service workflows, naming retailers and consumer brands as early examples (including Kroger, Lowe’s, Papa Johns, Woolworths).
A headline-grabbing signal: Walmart is integrating shopping directly into Gemini’s chat experience—moving beyond “recommendations” to cart-building and purchase completion inside an AI assistant interface.

3) The market data says OpenAI still leads enterprise spending—but the field is fluid

Corporate spend signals (while imperfect) show OpenAI regaining momentum in enterprise usage, with competitors also growing—suggesting many companies are adopting multiple model providers depending on use case, cost, and governance.
Meanwhile, broad surveys indicate AI usage is becoming near-universal at least in pockets of the org—though scaling remains uneven.

4) Platform alliances are turning rivals into “frenemies”

A defining feature of this cycle is coopetition: Google is simultaneously competing with OpenAI while also striking deals that place AI models into enterprise distribution channels and infrastructure ecosystems.

Strategic Outlook: The Author’s Analysis

The uncomfortable truth: “Agentic AI” is powerful—but still unreliable in the exact ways enterprises care about

Despite rapid gains, today’s agentic systems face four structural limitations:

1. Reliability under ambiguity
   Agents can look confident while making brittle assumptions, especially when tasks involve policy edge-cases (refund rules, medical claims, regulated workflows). A single wrong step can be costlier than a wrong answer.
2. Verification is not solved at scale
   Enterprises don’t just need outputs; they need auditability: why a decision was made, what data was used, whether permissions were respected, and how errors are caught before they ship to customers.
3. Security + data governance become the product
   As assistants gain the ability to act (not just suggest), identity, access control, and data boundaries become the differentiator. This is where “AI pilots” die—because legal/compliance teams can’t sign off without robust controls.
4. Economics are shifting from training bragging rights to inference ROI
   Leaders are increasingly acknowledging bubble-like dynamics in AI investment and the strain of compute/energy demands—meaning boards will demand measurable unit economics, not just capability demos.

The real potential: AI becomes a transaction layer—not just an interface layer

Walmart’s move is a tell: assistants are being positioned as the front door to commerce, compressing search, discovery, comparison, and checkout into a single conversational flow.
If this sticks, it rewires customer acquisition: brands will compete to become the default “agent-native” option (inventory, fulfillment, returns, support), and assistants will steer demand.

My prediction: the companies that will be left behind if they don’t pivot

Over the next 18–24 months, the laggards won’t be the firms without an AI demo—they’ll be the firms without agent-ready operations (APIs, identity, policies, inventory, fulfillment, customer service tooling).

If they don’t pivot aggressively, expect pressure on:

• Traditional contact-center outsourcers and legacy CX operators whose business model is “humans as the interface,” unless they rebuild around AI supervision + high-complexity exception handling.
• Mid-tier retailers and consumer brands that still treat digital as “website + app,” while leaders turn AI assistants into the primary shopping surface (Walmart is already moving).
• SaaS vendors with seat-based pricing and shallow workflows that bolt on chat but don’t enable end-to-end task completion; they’ll face churn as customers consolidate around platforms that can execute, not just answer.

Bottom line: Hassabis talking to Pichai “every day” is not a trivia detail—it’s a signal that Google believes the winner of the next phase won’t be the best model in isolation, but the company that ships the most trusted, governed, transaction-capable agents into real workflows at scale.