Kraken Bets on Agentic Trading to Survive the Bear Market

Kraken, founded in 2011 and one of the oldest surviving crypto exchanges, is rebuilding its app around agentic trading. The pitch from chief data officer Kamo Asatryan is direct: give everyday users the same engagement pattern that keeps professional traders active through bear markets, and do it through a conversational AI layer sitting on top of the exchange.

That is a bigger claim than it sounds. It reframes the exchange's product surface away from order books and toward an agent that talks, recommends, and executes on confirmation. My read: this is Kraken's answer to a structural retention problem the whole sector shares, and it lands during a prolonged crypto price downturn when experimentation is cheap and user acquisition is expensive.

Key Details

According to CNBC, which was told about the rebuild exclusively, Kraken's platform will give users access to agents capable of continuously monitoring markets, identifying investing opportunities, and executing trades in real time. Crucially, the experience is agentic but not fully autonomous. Trades and recommendations are only executed with the customer's explicit confirmation. That is a meaningful design choice, and I will come back to it.

The onboarding flow uses AI to capture users' goals, risk tolerance, funding preferences, and financial profile in a single streamlined path. From that, the AI builds a draft portfolio that users can review, adjust, and approve, with explanations attached to each recommendation. Post-investment, the system delivers AI-curated insights, portfolio-relevant news, and proactive recommendations, including flagging idle cash that could be optimized. Over time, both the conversation and the interface itself are meant to adapt to the individual user.

Asatryan framed the interaction model plainly: "Talking to Kraken should be like talking to your well-informed best friend who knows a lot about finance but also knows a lot about you." He argued that "in this new world, there's an opportunity for everyday people to become high-frequency traders and do so using plain English by just talking to their well informed best friend."

The business logic sits underneath the product logic. Kraken's core user base is institutions, trading firms, professional traders, and active use traders. Retail customers on the platform have typically been trading crypto for years. Asatryan noted the usual retail pattern elsewhere: "buy in at the peak, sell when prices are down, and they churn." Kraken is positioning agentic trading as the fix for that churn curve, alongside a broader expansion into payments, banking, and lending as it moves toward becoming a full stack financial services platform.

Why This Matters for Crypto and DeFi

Compare two numbers that aren't in the source but frame the problem: exchange revenue is roughly linear with trading volume, and trading volume for retail collapses in bear markets while pro-desk volume holds. Asatryan's own diagnosis captures it: "Traditionally, exchanges have trouble in bear markets because most of their customers have trouble in bear markets." If agentic tooling can flatten that curve for retail, even partially, the revenue implications are material. If it cannot, this is an expensive UX experiment.

The "confirmation required" design is the interesting technical decision. Fully autonomous agents executing on-exchange would immediately collide with US regulatory posture around discretionary trading, custody, and investment advice. Keeping a human confirmation step in the loop keeps the product on the near side of that line. The source does not disclose how Kraken handles the compliance and suitability framing of AI-generated recommendations, which matters because the SEC has been progressively tightening rules around algorithmic recommendation systems and conflicts of interest, as visible in ongoing SEC rulemaking. That is a testable bound: if the product ships in the US without a registered advisory wrapper, it must be structured as tool-assisted self-direction, not advice.

There is also a competitive read. Coinbase and Gemini have already introduced AI-assisted trading and developer tools. Kraken is not first, but it is the first to describe the rebuild as the app's organizing principle rather than a feature bolted onto an existing interface. The question is whether "agent-first" is a durable moat or a UX skin that any well-capitalized competitor can replicate in a quarter. My working assumption: the moat, if there is one, sits in the personalization data loop (goals, risk profile, historical behavior), not in the LLM itself.

Prediction: if this works, we should see Kraken's retail daily active user retention over a 90-day post-onboarding window improve versus the pre-launch cohort within two quarters of general availability. If that number doesn't move, the thesis is wrong.

Industry Impact

For engineering teams building exchange or brokerage infrastructure, the interesting layer isn't the chat interface. It's the plumbing: a system that can continuously monitor markets, generate recommendations, produce natural-language explanations, and route confirmed intents through the existing matching engine and risk stack, all under low enough latency that the "high-frequency traders in plain English" framing isn't marketing fiction.

That stack has several unknowns the source doesn't address. What is the recommendation latency budget? What is the fallback behavior when the model is uncertain or the market data feed is stale? How are hallucinated recommendations caught before they reach a user with a funded account? These aren't rhetorical questions. They are the difference between an agent that improves engagement and one that generates a regulatory or reputational incident on day one.

For DeFi teams watching this, the interesting divergence is between centralized agentic UX (Kraken's approach) and on-chain agent frameworks that would need to interact with smart contracts, oracles, and cross-chain messaging. An agent operating against DeFi rails needs deterministic price feeds and cross-chain execution primitives of the sort described in Chainlink's documentation, plus wallet-level intent authorization. Kraken's model sidesteps all of that by keeping execution inside its own venue. That is a faster path to shipping and a narrower feature set.

For fintech and iGaming platform leads, the transferable pattern is the onboarding-to-personalization pipeline: capture profile once, generate a proposed configuration, explain it, iterate. That structure applies well beyond crypto trading, and it is the piece most likely to be copied.

What to Watch

Three signals will tell us whether this is a genuine growth strategy or a bear market press cycle.

First, disclosure of activation metrics. If Kraken publishes any post-launch data on retail trade frequency, average session length, or 30-day retention versus the pre-agent baseline, watch whether the retail cohort actually behaves more like the pro cohort Asatryan is comparing them to. The source does not disclose current retention numbers, which matters because without a baseline any post-launch improvement claim is unfalsifiable.

Second, the regulatory framing at launch. Whether the agent's outputs are positioned as "recommendations," "insights," or "tools" will tell us how Kraken's legal team read the current rules. That framing is the single biggest constraint on how aggressive the product can be.

Third, the competitive response. If Coinbase and Gemini move from "AI-assisted trading and developer tools" to full agent-first app rebuilds within six months, the category has consolidated around this pattern. If they don't, Kraken is running an experiment the rest of the industry expects to fail.

Testable prediction: within 12 months, at least one major US crypto exchange will announce an agentic trading product with a materially similar confirmation-in-the-loop design. If none do, Kraken's bet is idiosyncratic rather than sector-wide.

Key Takeaways

• Kraken is rebuilding its app around agentic trading, with agents that monitor markets, recommend trades, and execute only on explicit user confirmation.
• The strategic target is retail retention through bear markets, addressing the "buy at peak, sell at bottom, churn" pattern Asatryan described.
• Coinbase and Gemini have shipped AI-assisted tools, but Kraken is the first to describe the rebuild as the app's central organizing principle rather than a feature.
• The confirmation-required design keeps the product on the near side of US regulatory lines around discretionary trading and investment advice.
• The unknown that matters most: whether retail engagement metrics actually converge toward the pro-trader pattern once the agents ship, or whether this remains a UX story without a retention story.