Coinbase's AI Strategy: How the Crypto Leader Is Rethinking Onboarding and Customer Discovery in 2026

TL;DR#

Coinbase's AI strategy in 2026 is among the most aggressive of any consumer financial company: the exchange has rebuilt internal compliance around AI agents to cut account-restriction resolution times by roughly 90%, now generates more than half of its daily code with AI, and is positioning itself as the payments rail for the agent economy through AgentKit, Agentic Wallets, and the x402 protocol — which had processed over 50 million transactions by late 2025. Yet the company that serves an estimated 120 million users and reported $2.8 billion in subscription and services revenue for full-year 2025 still learns why a new user abandons identity verification, gets spooked at the funding step, or distrusts a self-custody prompt the old way: drop-off funnels, support tickets, and survey scores. In a category where anxiety and trust are the real conversion levers, the missing signal isn't what users do — it's why they stop. This post maps Coinbase's real AI moves, shows where funnel-and-survey listening bottlenecks onboarding insight in a high-stakes vertical, and explains why conversational AI interviews capture the reasoning that drop-off analytics never will.

What is Coinbase's AI strategy in 2026?#

Coinbase's AI strategy in 2026 is a two-front program: using AI agents to compress its own compliance, support, and engineering operations, while building the financial infrastructure that lets other AI agents transact onchain. CEO Brian Armstrong has described the long-term direction as "intelligence, with humans around the edge" — smaller "AI-native pods" of people working alongside automated systems rather than large manual teams.

The internal results are concrete. Coinbase rebuilt nearly every compliance workflow around AI agents and cut account-restriction resolution times by about 90%, with AI systems now processing roughly 55% of U.S. fraud cases internally, according to Armstrong's May 2026 remarks reported by crypto.news. Armstrong has also said more than 50% of daily code at Coinbase is now AI-generated, after he publicly mandated AI-coding-assistant adoption and reportedly let go engineers who refused. The restructuring included roughly a 14% headcount reduction reorganized into those AI-native pods.

The external bet is bigger. Coinbase is trying to own the layer where autonomous agents pay for things, and that bet is the most ambitious piece of its 2026 AI posture.

Where Coinbase already uses AI: agents, payments, and the onchain machine economy#

Coinbase has shipped real AI infrastructure, not slideware — most of it aimed at letting software agents hold money and transact without a human in the loop. The flagship pieces are AgentKit, Agentic Wallets, and the x402 payments protocol.

• AgentKit gives developers a framework to embed onchain wallets and financial actions directly into AI agents, so an agent can spend, earn, and trade programmatically.
• Agentic Wallets is purpose-built wallet infrastructure for agents, with autonomous spending, earning, and trading plus security guardrails — described by Coinbase as the first wallet layer built specifically for agents.
• x402 is an open payments standard that repurposes the dormant HTTP 402 "Payment Required" status code so any API endpoint becomes a paywall a machine can pay instantly in stablecoins. By late 2025 it had been battle-tested across more than 50 million transactions, per Coinbase's developer documentation.

The ecosystem momentum is notable: Coinbase and Cloudflare co-founded the x402 Foundation in September 2025, with members that now include Google, Visa, AWS, Circle, Anthropic, and Vercel, and Coinbase has enabled agentic commerce for OpenAI's Agents SDK. This is the same structural argument Perspective AI makes about research — that the interface to value is shifting from static forms and dashboards to autonomous, conversational software. The pattern shows up across the consumer-fintech landscape, from Robinhood's customer-conversation strategy to how PayPal listens across 430 million accounts and Block and Square's seller ecosystem.

The Coinbase scale problem: 120 million users, every one a trust decision#

Coinbase operates at a scale where small onboarding-conversion differences are worth tens of millions of dollars, and where every signup is fundamentally a trust decision. Coinbase last disclosed roughly 108 million verified users at the end of 2024, and outside estimates put total monthly users near 120 million in 2025, per third-party analyses of its disclosures. In Q2 2025 it counted 8.7 million monthly transacting users — meaning only a fraction of the user base transacts in any given month, and re-activation is as much a relationship problem as an acquisition one.

The money behind those relationships is large. Coinbase reported $2.8 billion in subscription and services revenue for full-year 2025, up 23% year over year, and now has 12 products generating more than $100 million in annualized revenue each, according to its Q4 2025 shareholder letter filed with the SEC. Diversification into staking, stablecoins, custody, and the developer platform means Coinbase is no longer a single funnel — it's a portfolio of onboarding journeys, each with its own anxiety profile.

That's the crux. Crypto onboarding isn't a checkout flow; it's a sequence of trust hurdles — identity verification (KYC), bank or card funding, two-factor setup, the first trade, and the leap to self-custody. Each step is a place a user can quietly decide the product isn't for them. The deeper guidance on closing that loop lives in the complete guide to AI-powered customer experience from first touch to renewal and how to build a voice-of-customer program from scratch in 2026.

Why onboarding and trust are make-or-break in crypto#

In crypto, onboarding friction and trust deficits are the single largest controllable drag on growth — and the industry's drop-off numbers are brutal. Crypto KYC verification commonly sees 50–80% of started verifications abandoned, and across financial services roughly one in four users (about 25%) abandon onboarding specifically because of KYC friction, according to fintech onboarding research compiled by Datakeen. Slow or complex onboarding caused 70% of financial institutions to lose prospective clients in 2025, up from 67% the prior year — and abandoned KYC processes alone strip an estimated $3.3 billion annually from the banking sector.

A few precise data points frame the stakes for a company Coinbase's size:

• 50–80% crypto KYC drop-off means the verification step, not acquisition, is often the largest leak in the funnel.
• ~25% of financial-services users abandon onboarding because of KYC friction specifically.
• 70% of institutions lost prospects to slow onboarding in 2025 — friction is a churn event before the relationship even starts.

Coinbase's 90% reduction in restriction-resolution time directly attacks one of these hurdles. But here's the limit: that win was about resolving a restriction faster once a workflow flags it. It doesn't tell Coinbase why a user who never finished KYC walked away, what made a first-time buyer distrust the funding screen, or which unspoken fear ("is this a scam?", "will I lose my keys?") killed the self-custody upgrade. Speed fixes the mechanics. It does not capture meaning.

Where form-and-funnel listening bottlenecks onboarding insight#

The bottleneck is structural: drop-off funnels and post-signup surveys can tell Coinbase where users abandon, but never why — and in a high-anxiety category, the "why" is the entire game. This is the gap Perspective AI exists to close, and it's why AI-first customer research cannot start with a web form.

Consider the three dominant ways a company at Coinbase's scale "listens" to onboarding today, and where each breaks:

All three flatten a messy, emotional decision into a field or a number. Forms front-load effort before the user feels understood, and they fail precisely at uncertainty — the "I'm not sure this is safe" and "it depends what my bank thinks" moments that decide whether someone funds an account. That's the same limitation documented in AI vs surveys: why conversations win for real customer research and the case for an AI survey alternative that rethinks research without the survey pattern. The peer-fintech version of this same trust-and-funding problem shows up in Chime's move to replace onboarding forms, SoFi's member-first conversational financial discovery, and Affirm's merchant-and-customer onboarding discovery.

How conversational AI interviews capture the "why" behind abandonment and trust#

Conversational AI interviews capture onboarding insight that funnels and surveys structurally cannot, because they let the user explain the abandonment in their own words and follow up on the vague answer in real time. Instead of a one-way "Rate your verification experience 1–5," an AI interviewer agent asks "You started verifying your identity and didn't finish — walk me through what was going through your head," and then probes the answer: "You said it felt risky — risky how?"

That follow-up is the whole point. The highest-value onboarding signal — "I didn't trust where my ID photo was going," "my bank declined the transfer and I assumed Coinbase was the problem," "I didn't understand the difference between the exchange and the wallet" — only surfaces when something can ask a second question. A static form cannot. For Coinbase-scale volume, that has to happen across thousands of conversations at once, not a handful of moderated sessions, which is the core of running customer interviews at scale via the research workflow.

Three places this fits the Coinbase journey directly:

1. Post-abandonment KYC interviews. Trigger a conversational concierge agent when a user exits verification, asking what stopped them — capturing privacy fears, document confusion, and competitor comparisons that a funnel records only as a percentage.
2. Funding-failure recovery. When a first deposit fails or stalls, an interview distinguishes a technical decline from a trust collapse — two failures that look identical in analytics but need opposite fixes.
3. Trust and self-custody discovery. Before shipping a new wallet or self-custody prompt, interview real users about what "owning your keys" makes them feel, surfacing the anxieties a product team needs to design around. The customer-research stack that supports this is mapped in the customer research tools modern product and CX teams actually use in 2026 and the complete guide to voice-of-customer programs in 2026.

The teams who own these moments — CX teams defending activation and support, product teams shaping the funding and custody flows — get reasoning they can act on, not a dashboard that says "23% dropped at step 3." Other consumer-fintech case studies follow the same arc, including Brex's conversational discovery for startup banking and Mercury's onboarding research.

The strategic gap in Coinbase's AI strategy#

Coinbase has world-class AI on the operational and infrastructure sides — agents that resolve restrictions, generate code, and move money — but the listening side of its AI strategy still runs on the survey-era stack. The company that is building rails for autonomous agents to converse with APIs is, for its own users, mostly still reading funnels and scores. That asymmetry is the opportunity.

The same conversational interface Coinbase is betting the agent economy on is exactly what its onboarding research needs: a system that asks, follows up, and captures meaning rather than fields. Closing that gap doesn't require a new compliance overhaul — it requires treating the "why behind abandonment" as a first-class research input, gathered through conversation at the same scale Coinbase does everything else. The broader buyer framing for that tooling is laid out in the buyer's framework for AI customer engagement software in 2026 and the best AI customer-insight platforms for enterprise in 2026.

Frequently Asked Questions#

What is Coinbase's AI strategy?#

Coinbase's AI strategy has two halves: applying AI agents to its own operations and building infrastructure for the broader agent economy. Internally, Coinbase rebuilt compliance around AI to cut restriction-resolution times by about 90% and now generates over half its daily code with AI. Externally, it offers AgentKit, Agentic Wallets, and the x402 payments protocol so autonomous AI agents can transact onchain. CEO Brian Armstrong frames the direction as "intelligence, with humans around the edge."

What are Coinbase AI agents and AgentKit?#

Coinbase AI agents are software agents that can hold a wallet and transact onchain autonomously, built using Coinbase's AgentKit framework. AgentKit lets developers embed onchain wallets and financial actions — spending, earning, trading — directly into AI agents. Coinbase's related Agentic Wallets product adds purpose-built wallet infrastructure with security guardrails, and the x402 protocol provides the machine-to-machine payment rail, having processed more than 50 million transactions by late 2025.

Why is onboarding so hard for crypto platforms like Coinbase?#

Crypto onboarding is hard because it strings together multiple high-anxiety trust hurdles — identity verification, account funding, security setup, and self-custody — that any one of which can cause silent abandonment. Crypto KYC commonly sees 50–80% of started verifications abandoned, and about 25% of financial-services users abandon onboarding due to KYC friction. Each step is both a compliance requirement and a moment where a user can decide they don't trust the product.

How can Coinbase learn why users abandon onboarding?#

Coinbase can learn why users abandon onboarding by running conversational AI interviews at the moment of drop-off, instead of relying only on funnel analytics and surveys. Funnels show where users exit but not why; surveys reduce reasons to scores. An AI interviewer can ask an abandoning user what stopped them and follow up on vague answers in real time — surfacing privacy fears, funding confusion, or distrust that quantitative tools never capture, across thousands of conversations at once.

What's the difference between Coinbase's operational AI and customer-research AI?#

Coinbase's operational AI automates internal work — compliance, fraud, code generation — while customer-research AI would capture the reasoning behind user behavior. Coinbase has advanced operational AI (90% faster restriction resolution, 55% of U.S. fraud cases handled by AI) but still studies onboarding and trust mostly through drop-off funnels and surveys. Conversational AI interviews close that gap by gathering the "why" behind abandonment, the layer operational metrics leave blank.

Conclusion: the missing half of Coinbase's AI strategy#

Coinbase's AI strategy is a genuine leader's playbook — AI-native compliance, agent-built code, and an onchain payments protocol positioning the company as infrastructure for the agent economy. But for a business where 120 million users each make a trust decision and crypto KYC drop-off runs as high as 80%, the most valuable signal in the business — why a user abandons verification, distrusts funding, or hesitates at self-custody — still arrives as a percentage on a funnel, not a reason in a transcript. Operational AI made Coinbase faster. The next gain is qualitative: understanding the anxiety behind the abandonment.

That's the half of the AI strategy forms and surveys can't deliver, and it's exactly what Perspective AI is built for — conversational AI interviews that ask, follow up, and capture the "why" at the scale a company like Coinbase operates. Start a research study or explore the AI interviewer agent to hear the reasoning behind your onboarding numbers, not just the numbers.