| The Agentic Enterprise |
AK · Wed, Jul 1, 2026 · 8 min |
|
|
Wednesday, July 1, 2026
The data layer is the new control plane.
Snowflake and Databricks both branded themselves the control plane for the agentic enterprise. The model just became the swappable part.
Your real lock-in decision moved one layer down. It is no longer which model. It is whose platform runs and governs your agents.
|
|
The Lead
Within a few weeks of each other, Snowflake and Databricks made the same move.
Snowflake shipped xAI's Grok into Cortex as generally available, dropping it beside Claude, GPT, Gemini, Llama, Mistral, and DeepSeek, and put out a press release calling itself “the control plane for the agentic enterprise.” Databricks reframed its lakehouse in exactly those terms, launched a governance layer over every model and agent, and made its agent platform model-neutral by default. Two of the largest enterprise data companies placed an identical bet: own the governed data, and let the model become the part you swap. For a CIO, that quietly rewrites the most important AI decision of the year. It is no longer which model. It is whose platform runs and governs your agents.
|
|
The Big Story
Infrastructure
|
The model is the commodity. The data platform is the moat.
| S |
nowflake and Databricks spent June making the same argument from two directions. Snowflake made Grok generally available in Cortex, alongside Claude, GPT, Gemini, Llama, Mistral, and DeepSeek, and issued a release naming itself the control plane for the agentic enterprise. Databricks used its summit to reframe the lakehouse the same way, launched Unity AI Gateway to govern every model, MCP, and external agent from one place, and made Agent Bricks model-neutral by default. The message from both: the model is interchangeable, the governed data underneath it is not. |
|
Swapping a model is a config change. Swapping your governed data platform is a migration. That is where the new lock-in lives.
|
The logic holds. An agent is only as good as the data and permissions it runs against, and that context lives in the warehouse, not the model. Whoever governs the data, the identity, the cost controls, and the audit trail governs the agents. The frontier labs spent three years making the model the center of gravity. The data platforms are quietly pulling it back to themselves.
For enterprise buyers, the strategic question flips. It is no longer which model do we standardize on — a choice that just got cheap and reversible. It is which platform runs and governs our agents, a choice that is expensive and sticky. Model neutrality is becoming the default architecture precisely because it makes the platform, not the lab, the thing you cannot easily leave.
|
The Spearhead Take
Choose your control plane on purpose, not by inertia. Keep models behind an abstraction so you can swap them as price and capability move. But treat the data-and-governance platform as the real commitment, because it is, and negotiate it like the multi-year decision it actually is.
|
|
|
The Obvious & The Overlooked
What the control-plane land grab says, and what it quietly admits.
|
The Obvious
Enterprises want model optionality.
Snowflake now serves Grok, Claude, GPT, Gemini, Llama, Mistral, and DeepSeek from one place. Flexera
Data platforms are climbing the stack.
Databricks reframed the lakehouse as the agent control plane at its annual summit. Bain
Agents are flooding into production.
Gartner expects 40% of enterprise apps to embed them by year-end, up from under 5%. InformationWeek
|
The Overlooked
The lock-in moved to the data layer.
Swapping a model is easy; swapping the governed data platform is not. SiliconANGLE
Model neutrality disintermediates the labs.
When any model plugs in, no single lab owns the customer relationship. Flexera
Governance is the product now, not a feature.
Databricks' Unity AI Gateway governs every model, MCP, and external agent from one point. Bain
The “control plane” label is doing strategic work.
Both vendors chose the same phrase in the same month to claim the same ground. Snowflake
|
|
|
Moving Pieces
Five developments worth a CIO's attention.
Product
GPT-5.6 Sol heads to Cerebras at 750 tokens per second
OpenAI will run GPT-5.6 Sol on Cerebras hardware at up to 750 tokens per second in July, initially for select customers as capacity scales. Sol, previewed June 26 alongside the cheaper Terra and Luna, remains gated to roughly 20 partner organizations individually approved by the US government, with general availability promised in the coming weeks. The signal for buyers is that raw speed is becoming a distinct, purchasable frontier. For long-horizon agentic work, latency is a cost and a capability at once, and the vendors are starting to price it that way.
Infrastructure
Baseten raises $1.5B, and it is not a model lab
Baseten closed a $1.5 billion Series F led by Altimeter, Conviction, and Spark at up to a $13 billion valuation, on roughly 20x year-over-year revenue growth and more than a billion inference requests a day. It was the largest US venture round of June, and it builds none of its own models. TensorWave, Cyera, and others pushed AI-infrastructure raises past $2 billion for the month. The pattern reinforces today's theme: capital is accruing to the layers that run and govern models in production — the serving and security plumbing — not to the models themselves.
Deals
Taktile raises $110M to put AI decisioning inside regulated finance
Taktile, a New York decisioning platform, closed a $110 million Series C led by Goldman Sachs to expand AI-driven decisions for banks and lenders. The target is one of the harder enterprise problems: using models where an error creates credit, compliance, and legal exposure, not just a bad answer. A Tier-1 bank leading the round is the signal. Regulated finance is done experimenting with chat and is now buying infrastructure to let models make governed, auditable decisions, which is exactly the use case most vendors avoid because it is the hardest to get right.
Workforce
Oracle books 21,000 cuts as AI becomes the year's most-cited reason
Oracle disclosed 21,000 job cuts over twelve months in its June 22 annual filing, part of a tech sector that has shed roughly 123,000 roles this year with AI now the single most-cited reason. The honest caveat is real: analysts warn of “AI washing,” where overhiring or weak revenue gets rebranded as an efficiency story for investors. For enterprise leaders, the useful read is not the headline count but the pattern. Companies are restructuring around the assumption that AI changes headcount math, whether or not the productivity data yet supports the size of the cut.
Policy
The EU AI Act blinks, pushing high-risk rules to late 2027
The European Parliament adopted the Digital Omnibus on June 16, extending high-risk obligations — including biometrics, critical infrastructure, employment, and education — to December 2027. Chatbot transparency rules still land in August 2026, and content-labeling requirements in December 2026. For US multinationals, the delay is breathing room, not a reprieve. The near-term compliance work arrives on schedule this year, while the heavier high-risk regime gives deployers eighteen more months to build governance that most do not yet have.
|
|
On the Radar
Nine signals, sharpened.
| Economics |
Microsoft’s July 1 pricing changes take effect today. Base Microsoft 365 prices rise, enterprise Copilot volume discounts expired June 30, and Security Copilot moves to metered compute units at $6 over the monthly cap. US Cloud |
| Compute |
Groq closed $650M for its inference cloud. Infinitum and Disruptive led the round for the custom-chip inference challenger, another bet that fast, cheap serving is the contested layer. Crunchbase |
| Product |
GPT-5.6 general availability is promised “in coming weeks.” OpenAI says Sol, Terra, and Luna will reach ChatGPT, Codex, and the API broadly, but has committed to no firm date. OpenAI |
| Deployment |
ServiceNow and Accenture launched a Forward Deployed Engineering program. Embedded engineers will stand up agentic AI inside clients, mirroring the consulting-led playbook OpenAI is also using. Accenture |
| Product |
Microsoft Foundry’s Hosted Agents reached general availability. The service adds built-in security controls, adaptive evaluations, and one-click publishing to Teams and Microsoft 365 Copilot. Microsoft |
| Deals |
General Intuition raised $320M Series A at a $2.3B valuation. Khosla led the round for a foundation model trained on gameplay, with Jeff Bezos and General Catalyst joining. Crunchbase |
| Policy |
EU chatbot transparency rules take effect in August 2026. Deployers must disclose when users are interacting with AI, the first Digital Omnibus obligation to bite this year. artificialintelligenceact.eu |
| Security |
OpenAI staggered GPT-5.6 at the US government’s request. Access was limited to about 20 vetted partners pending a national-security review, a new friction point in enterprise model access. VentureBeat |
| Deals |
Cyera raised $600M Series F for AI data security, lifting total funding to roughly $1.2 billion as governance spend rises with agent adoption. Tech Startups |
|
|
Quick Hits
The wider field, one line each.
| TensorWave raised a $350M Series B, an all-AMD AI cloud co-led by AMD Ventures at a $1.55B valuation. Tech Startups |
| Assort Health raised a $120M Series C for healthcare voice AI, taking total funding past $222M. Tech Startups |
| Mirendil raised a $200M seed led by Andreessen Horowitz and Kleiner Perkins to build an AI-for-R&D frontier lab. Crunchbase |
| XCures landed a $46M Series B to clean up messy medical records with AI. Crunchbase |
| Databricks launched Agent Bricks, putting frontier and open models in one governed platform. Databricks |
| Databricks and the Linux Foundation launched OpenSharing, an open protocol for sharing agent skills and models across platforms. Bain |
| Snowflake Cortex Code now reaches external systems including AWS Glue, Databricks, and Postgres. Snowflake |
| Micron signed a strategic agreement with Anthropic covering memory architecture and an investment in Anthropic’s Series H. Micron |
| Anthropic opened a Seoul office and announced partnerships across the Korean AI ecosystem. Anthropic |
| NVIDIA and ServiceNow partnered on autonomous enterprise agents built on ServiceNow’s platform. NVIDIA |
| Intuit will cut roughly 3,000 jobs, about 17% of staff, in a restructuring toward AI. Forbes |
| Cisco is cutting about 4,000 jobs as it pivots toward AI-driven networking and security. Forbes |
|
|
The Number
40%
The share of enterprise applications Gartner expects to embed task-specific AI agents by the end of 2026, up from under 5% a year ago.
That is one of the steepest adoption curves any enterprise software category has posted, and it is exactly why the control-plane land grab is happening now. When agents flood into production this fast, whoever governs where they run captures the account.
|
|
Counter-Signal
Governance
The control plane is lock-in wearing a better outfit.
The clean story is that model neutrality sets enterprises free. Any model plugs in, no lab owns you, switching is a config change. True, as far as it goes.
Look at what you traded. The model got commoditized; the data platform got indispensable. Consolidating your agents, governance, identity, and audit trail onto one vendor’s control plane is not the end of lock-in. It is lock-in relocated to the layer with the highest switching costs in the whole stack. A model migration is a weekend. A data-platform migration is a multi-year program most companies never finish. Snowflake and Databricks are not offering optionality out of generosity. They are offering it because the optionality sits at the cheap layer and the stickiness sits at theirs. The answer is not to refuse a control plane; every serious agent program needs one. It is to enter that choice clear-eyed, and to keep your data portable enough that the threat to leave stays credible.
|
|
From the Field
The second half of the year starts with a quieter shift than the headlines suggest.
For three years the enterprise AI question was which model. Which lab, which benchmark, which frontier. This month two data companies answered it by making the question smaller. When Snowflake and Databricks both call themselves the control plane for the agentic enterprise within a few weeks, they are telling you where they think the value settled. Not in the model. In the governed place the model runs.
This is how infrastructure markets usually mature. The exciting layer commoditizes, and the boring layer underneath — the one holding the data, the permissions, and the audit trail — becomes the thing you cannot leave. The enterprises we work with feel this already. They have stopped agonizing over model choice and started treating it as a setting you can change. Their real deliberation now goes into the platform, because that is the decision they will live inside for years.
It is less exciting than a new frontier model and far more consequential. The model was never going to be the moat. The data always was.
Let’s get to production,
AK
|
|
|
|
Anthropic is a Spearhead technology partner and its Claude model produced this edition under human editorial direction. This edition does not center Anthropic favorably: Claude appears as one interchangeable model among seven in Snowflake’s Cortex roster — the explicit point being that the model layer, Claude included, is being commoditized beneath the data platform. Anthropic otherwise appears as a commercial actor (Micron strategic agreement, Seoul office). The same analytical frame was applied to OpenAI, Google, Microsoft, Databricks, and Snowflake. The Baseten figures ($1.5B raise, up to $13B valuation, ~20x revenue growth, 1B+ daily requests) are from Crunchbase; Oracle’s 21,000 cuts and the ~123,000 tech-sector figure are from Forbes and TechCrunch. The Gartner 40% projection is via InformationWeek. Microsoft’s July 1 pricing mechanics are from US Cloud. Several radar and quick-hit items rely on VC roundup aggregation pending primary confirmation. All editorial decisions are human-directed.
|
|
The Agentic Enterprise
Know more about AI than 95% of your peers. By 7 AM.
A daily AI intelligence briefing for enterprise leaders, published by Spearhead. We build AI systems that work. Strategy. Engineering. Production. Outcomes.
© 2026 Spearhead. All rights reserved.
|
|
