No AI Strategy Without a Data Strategy: Why Skills Are the New Apps and Intelligence Must Live Where Data Lives

February 18, 2026

No AI Strategy Without a Data Strategy

Why skills are the new apps, the terminal era of AI UX is ending, and intelligence must live where data lives

The enterprise AI conversation has a blind spot. Everyone is racing to deploy agents, fine-tune models, and build copilots — but almost no one is asking the question that determines whether any of it works: where does your data actually live, and is it ready for AI?

In a recent Masters of Automation episode, Baris Gultekin — Vice President of AI at Snowflake and co-creator of Google Now — makes the case that the next era of AI won't be defined by model size or benchmark scores. It will be defined by data gravity: the architectural reality that intelligence must move to where data already lives, not the other way around.

It's a deceptively simple idea that rewires how you think about governance, trust, agent design, and the entire economics of intelligence.

"There is no AI strategy without a data strategy. Ultimately, you still need to think about how you bring all of your data together so that your data assets can be joined, can be combined when you're trying to bring insights across your data." — Baris Gultekin, VP of AI at Snowflake

Watch/Listen

The hard lesson from the assistant era

Before leading Snowflake's AI product portfolio, Baris spent 16 years at Google. He co-created Google Now — the first AI assistant that predicted what you needed before you asked — and helped grow Google Assistant from 10 million to 500 million monthly users.

The experience left him with a scar that shapes everything he builds today.

The problem wasn't ambition. It was architecture. Every use case had to be hand-built: when someone says this, do that. It was brittle by nature. And when you wrap that brittleness in a natural-language interface, you create a devastating mismatch — users expect human-like intelligence, and when it works for two use cases but fails for twenty, trust collapses completely.

"When you make something more natural, there are expectations of human-like intelligence. When it works in a couple of your use cases but not most, that expectation breaks down. That was the biggest challenge with the Google Assistant era." — Baris Gultekin

Today's reasoning-capable models have closed that gap. But the lesson endures: naturalness creates expectations, and expectations demand reliability. The companies that win the agentic era won't just have the best models — they'll have the most trustworthy data foundations underneath them.

Bring AI to the data — not data to the AI

Snowflake's core thesis sounds obvious once you hear it, but most enterprises are still doing the opposite: they extract massive amounts of data, ship it to a separate AI platform, and hope the governance follows.

It doesn't.

When you bring AI to run next to data instead, something powerful happens. All of the access policies, governance rules, and security boundaries that already exist on your data platform are inherited automatically. You don't have to replicate them. You don't have to hope a third-party AI vendor respects them. The architecture enforces them.

But governance inheritance is just the beginning. By being close to data, Snowflake helps customers build what Baris calls AI-ready data — bringing the semantics of a business onto the data platform so that AI understands how data is structured, what metrics matter, what synonyms are used internally, and how different datasets relate to each other.

This is the unglamorous work that makes agents actually useful. An agent without business context is just an expensive autocomplete. An agent grounded in a company's semantic layer — its ontology of knowledge — can answer questions that previously required a team of analysts and a two-week turnaround.

Skills are the new apps

Here is where the conversation takes a sharp turn toward the future.

Baris describes skills — modular instructions and scripts that agents can load on demand — as the most exciting frontier in enterprise AI. The analogy he reaches for is The Matrix.

"These agents, it's like in The Matrix — 'I want to know karate' and I know karate. Being able to tap into these skills and then use it for various purposes expands the surface on which you can take action." — Baris Gultekin

Think about what this means. Instead of building monolithic applications with fixed UIs and rigid workflows, you create discrete skills: pull up contract data, draft a marketing campaign, summarize customer reviews, check renewal status. Each skill can embed deep business logic — the niche, proprietary knowledge that makes a company's processes different from everyone else's.

An agent loads a skill the way Neo downloads kung fu. Instantly. Composably. And the surface of what it can do expands exponentially with each new skill added.

Snowflake already built an internal agent for its own sales team that can look up any customer's information, surface open issues, check renewal timelines, and identify active use cases. But the power isn't in what it does today — it's in the fact that every new skill multiplies its capability without rebuilding anything.

This is a fundamentally different model from the application era. Applications are static. Skills are composable. Applications require users to learn interfaces. Skills require agents to load instructions. The shift from apps to skills is as significant as the shift from desktop software to SaaS — and it's happening faster.

The $500 billion SaaS question

If skills replace applications, what happens to Salesforce, SAP, and Workday?

Baris's answer is nuanced: the vendors don't disappear, but the interface layer transforms completely. The majority of interactions with these platforms will be driven by agents rather than humans. You won't click through Salesforce — your agent will talk to Salesforce's agent.

This is not a distant prediction. When natural language becomes the universal interface, the barriers between siloed enterprise systems start to dissolve. An agent doesn't care whether data lives in Salesforce, SAP, or a custom database — as long as there's a protocol to reach it.

And protocols are arriving fast. MCP handles context retrieval. New agent-to-agent protocols enable negotiation and commerce between systems. Because everything operates in natural language, Baris argues adoption will be faster than any prior protocol shift in enterprise history.

The implication for builders: the competitive moat is shifting from UI design to skill design. The companies that create the richest, most reliable skills — grounded in the deepest business logic — will own the agentic interface to their domains.

We are living in the terminal era of AI

One of the most provocative observations in the conversation is Baris's claim that we've gone back to the terminal.

Think about it. The dominant AI interface today is a text box. You type a prompt, you get a response. It's powerful — but it's the command line of the AI age. And just as the command line gave way to graphical user interfaces, and GUIs gave way to mobile, the text-box era will give way to something much richer.

Baris sees three modalities converging:

  • Visual interfaces that pack far more information density than text — forms, dashboards, interactive elements that let you tap instead of type
  • Voice that has already made a comeback and is getting more natural every month — brainstorming while driving, hands-free interactions
  • Multi-modal experiences that combine vision, voice, and text fluidly based on context

This matters because interface shapes adoption. Every time an interaction becomes more natural, usage explodes. The terminal-to-GUI transition created the personal computing revolution. The GUI-to-mobile transition created the smartphone revolution. The text-to-multimodal transition will create the agent revolution.

The companies building for a text-only future are building for the current terminal era. The ones designing for visual, voice, and multi-modal agent experiences are building for what comes next.

The oracle problem: trust when agents negotiate

When your agent negotiates with my agent, who holds the source of truth?

Baris frames this with elegant simplicity: each agent defends its master's outcome. Your agent knows your preferences, your budget, your constraints. My agent knows mine. Both are grounded in world knowledge — the web, market data, public information — but personalized with private context that makes them fundamentally different from each other.

There's no need for a single oracle. The common ground is the world's accessible information. The differentiation is in the private knowledge each agent carries on behalf of its user. This mirrors how human negotiation works — shared context, divergent interests, convergent resolution.

But this only works if the underlying data is trustworthy. If an agent pulls from stale data, incomplete records, or ungoverned sources, the negotiation breaks down. Trust in agents flows directly from trust in data. This is another reason why "no AI strategy without a data strategy" isn't a slogan — it's an architectural requirement.

The economics of intelligence

We've entered an era where intelligence itself is a commodity you can rent. Training a frontier model costs hundreds of millions. Inference at scale costs real money. But two forces are making intelligence radically more accessible:

Models are getting both better and cheaper — a combination that rarely happens in technology. The cost per token is dropping while capability per token is rising.

Natural language is democratizing data access. Analysts, executives, and operators who never wrote SQL can now query massive datasets in plain English. The barrier to insight has collapsed.

Baris sees the result as a growing pie, not a zero-sum fight between layers. Value is accruing at every level of the stack — hardware (NVIDIA), models (frontier labs), data platforms (Snowflake), and applications. The tide is lifting all boats because the total addressable opportunity keeps expanding.

For enterprise leaders, the strategic implication is clear: don't wait for the stack to settle before investing. The companies that build their data foundations now — breaking down silos, governing access, building semantic layers — will be positioned to capture value regardless of which model or application layer wins.

Trust is the gating factor for autonomy

Why aren't fully autonomous agents running end-to-end enterprise workflows yet?

It's not a capability problem. The technology is ready. It's a trust problem.

Baris draws a clean line: low-stakes automations are already running without humans in the loop. Customer support tickets get resolved. Data gets summarized. Reports get generated. For these, the error rate is acceptable and the cost of failure is low.

High-stakes decisions — financial approvals, legal actions, strategic choices — still have humans in the loop. But the efficiency gain is enormous: what used to take weeks now takes minutes, because the agent does the heavy lifting and the human provides the judgment.

The path to full autonomy isn't a single leap. It's a gradual expansion of the trust boundary as quality improves, evaluation datasets mature, and guardrails prove reliable. Baris recommends scoped agents — purpose-built for specific tasks with clear skill definitions, comprehensive test suites, and known failure modes.

This is the opposite of the "build one god agent" approach. It's modular, testable, and trustworthy by design.

Getting AI-ready: the enterprise playbook

Baris lays out a clear sequence for enterprises that want to be AI-ready:

  1. Break down data silos. Get all your data into one governed platform so it can be joined, combined, and queried across domains.
  2. Secure and govern. Ensure only the right people — and the right agents — have access to the right data. Inherit governance from the platform rather than bolting it on.
  3. Build the semantic layer. Define your metrics, your business terms, your synonyms, your ontology. This is what lets AI understand your business, not just your data.
  4. Make your knowledge base accessible. Documents, contracts, policies, reviews — all of it needs to be queryable and retrievable.
  5. Then build agents and skills. With the foundation in place, agents become dramatically more capable, more trustworthy, and more valuable.

The order matters. Companies that skip to step five — building agents on top of fragmented, ungoverned data — are building on sand.

A closing thought

The Google Assistant era proved that natural interfaces without deep intelligence create disappointment. The ChatGPT era is proving that deep intelligence without enterprise data creates toys. The next era — the one Baris is building toward at Snowflake — brings them together: reasoning-capable agents, grounded in governed enterprise data, armed with composable skills, operating across every modality.

The companies that get there first won't be the ones with the biggest models. They'll be the ones with the cleanest data, the richest semantic layers, and the most thoughtfully scoped agents.

There is no AI strategy without a data strategy. And the time to build that strategy was yesterday.

Masters of Automation explores the intersection of artificial intelligence, human autonomy, and the critical choices facing civilization. We believe the future remains unwritten, though it approaches with unprecedented speed.

Guest: Baris Gultekin

Host: Alp Uguray

Mentioned in this episode

Key chapters

  • 03:01 — From Google Now to Google Assistant: why the technology wasn't ready then
  • 05:24 — Bring AI to the data: Snowflake's core architecture thesis
  • 11:27 — Agent-to-agent protocols: MCP, commerce, and the agentic web
  • 13:25 — The Matrix moment: skills as the breakthrough for enterprise agents
  • 16:21 — We're back to the terminal: why AI UX is about to change
  • 20:07 — The economics of intelligence: democratization and the growing pie
  • 37:07 — Trust as the gating factor for agent autonomy
  • 44:14 — No AI strategy without a data strategy
  • 46:50 — Startup advice: passion, depth, and "you cannot not do it"