Mentioned in This Article
1. Coinbase AI agents plan rests on a 99.9% forecast
Coinbase CEO Brian Armstrong said on Tuesday, November 17, 2026, that artificial intelligence agents will account for "well over 99.9%" of all transactions within a decade, and that his company is building the accounts, payment tools and marketplaces those agents will need. Speaking at the TomorrowX conference in Austin, Texas, he also said bitcoin has likely hit bottom.
The prediction is a statement about volume rather than value. Armstrong's argument is that software agents acting on behalf of people and businesses will soon be far more numerous than humans, and that each of them will need a way to pay other agents for goods and services. If that happens, he said, it could change how money moves through the global economy.
For Coinbase, the forecast doubles as a business plan. The exchange is wagering that crypto will become the payment rail for that machine-driven economy, and that supplying it will produce revenue that does not rise and fall with the number of people buying and selling digital assets.
2. More agents than people, making smaller payments
Armstrong's reasoning starts from a simple count. There will be many more agents than there are humans, he said, and the payments they make to one another are likely to look different from the ones people make today. He expects agents to transact in smaller amounts and in real time, settling as work is done rather than through periodic invoices or monthly bills.
That is how a figure as high as 99.9% becomes plausible in his telling. A person might make a handful of purchases in a day. A piece of software carrying out a task could make hundreds of tiny payments in the course of completing it, each one to a different service. Multiply that across a population of agents larger than the human one and the share of transactions initiated by people shrinks to a rounding error, even if humans still control most of the money.
An AI agent, in this context, is software that can be given a goal and then carry out the steps needed to reach it, including choosing which tools and services to use along the way. Unlike a chatbot that answers a question and stops, an agent acts. Once it is acting in the world, it runs into the same problem any business does: the things it needs cost money, and the suppliers want to be paid.
The forecast is Armstrong's own view of where the technology is heading, not a measurement of current activity. It does, however, explain where Coinbase is putting its development effort.
3. How would AI agents pay each other?
Coinbase's answer is stablecoins moving over a blockchain, with no human signing off on each payment. The company is building financial accounts for agents, tools that let them send and receive funds, and marketplaces where they can find and hire one another. The goal is to let agents collaborate and pay for services without a person approving every transaction.
Armstrong illustrated the idea with a business that uses AI to produce an advertisement. Rather than relying on a single model to do everything, an agent could hire separate specialist services, one for the voiceover and another for the video, and pay each of them as the work is delivered. The business owner sets the task and the budget; the agent handles the procurement and the payments.
Payments of that kind may be too small or too frequent for traditional card networks to handle. Card systems were designed around purchases made by people, with fees and settlement times that make sense for a restaurant bill or an online order but not for a stream of payments worth fractions of a dollar. Armstrong said much of the early agent activity is running on Base, the blockchain Coinbase developed, and is denominated in USDC, the dollar-backed stablecoin.
Stablecoins suit the task for practical reasons. Their value is pegged to the dollar, so an agent paying for a service does not have to manage the price swings of an asset such as bitcoin. A transfer can be sent at any hour, to any address, for any amount, and software can hold and move the tokens directly. An agent cannot easily open a bank account or be issued a card in its own right, but it can be given a wallet with a defined balance and a defined set of permissions.
4. A search for revenue that does not depend on trading
Behind the agent strategy sits a long-standing weakness in Coinbase's business model. Trading fees remain the core of what the company earns, and that income depends heavily on crypto prices and on how active investors are. When markets cool, volumes fall and so does revenue.
The company's earnings show that the reliance on trading fees continues. Income from stablecoins, staking and custody has grown enough to offset part of the decline when markets slow, but not to replace it. Payments made by AI agents would be a different kind of revenue altogether, one tied to how much economic activity software carries out rather than to how many people are buying or selling bitcoin on a given day.
The timing of Armstrong's remarks underlines the point. He was speaking during a selloff, the kind of period in which a trading-led business is most exposed, and he used the occasion to describe a line of business whose growth would be driven by the spread of AI rather than by the mood of crypto investors.
Coinbase has already moved some distance from its origins as a venue for spot crypto trades. Prediction markets and derivatives now provide about 25% of its trading fee revenue, a sign that the mix of what customers trade on the platform has broadened considerably.
5. Rivals are crowding into the same businesses
That diversification is partly a response to competition arriving from several directions at once. Robinhood is expanding its crypto business and adding products that let customers trade stocks, options and digital assets inside a single app. Kraken and Binance compete with Coinbase directly, particularly in derivatives.
Prediction markets have become another contested area. Polymarket and Kalshi attract traders who want to bet on events such as elections and economic data releases, and both Robinhood and Coinbase are moving into that territory. The result is that the lines between a crypto exchange, a retail brokerage and an event-betting platform are blurring, as trading firms, brokerages and prediction platforms enter each other's markets.
In that environment, a payments business serving AI agents would give Coinbase something its competitors in trading do not obviously have. It would also build on assets the company already controls, namely the Base network and its close association with USDC.
6. Armstrong runs his own trading agents, with mixed results
The Coinbase chief is testing the idea with his own money. He said he runs "three or four" AI agents, each managing a different portfolio, and described the outcome so far as mixed: some of the portfolios are up and some are down.
Coinbase is turning that experiment into a product. The company is developing tools that would let users build trading strategies, publish the results, and allow other customers to copy them. A user who trusted a particular strategy could, in effect, let it trade on their behalf.
Armstrong stressed that the tools will come with safeguards. He pointed to spending limits, account permissions that restrict what an agent is allowed to do, and the ability to test a strategy before putting real funds at risk. Those controls address the most obvious concern with handing money to software, which is that an agent acting without supervision could lose it quickly.
His own mixed record is a useful caution alongside the larger forecast. An agent that can place trades is not thereby an agent that places good ones, and the fact that some of the chief executive's portfolios are down shows that automated strategies carry the same market risk as any other. The copy-trading tools would make results public, which lets users compare strategies before following one.
7. Why is bitcoin dropping? Armstrong points to big tech listings
Armstrong's explanation for the recent selloff is that investors have been pulling money out of bitcoin to buy into major technology listings. Bitcoin (BTC) was trading at roughly $63,700 as he spoke, and he placed part of the blame on that rotation rather than on any change in the asset's prospects.
He compared the current episode with an earlier one, when investors moved out of bitcoin ahead of SpaceX's initial public offering and then returned to crypto months later. He suggested that Anthropic's IPO could be driving a similar shift now. The order books for that offering were set to close at 4 p.m. Eastern on Tuesday, the same day he made the remarks.
If that reading is correct, the pressure on bitcoin is temporary and tied to a specific event. Money that left to take part in a large listing would be expected to come back once the shares are allocated and trading. That is Armstrong's interpretation of investor flows, and he offered it as a possible explanation rather than an established cause.
8. The $400,000 target for 2030 stays in place
Despite the decline, Armstrong said he remains optimistic and is standing by his forecast that bitcoin could reach $400,000 by 2030, a level he called "pretty reasonable." Reaching it from current prices would require the asset to rise more than sixfold in a little over four years.
His confidence rests on bitcoin's four-year cycle, the pattern in which the market has historically moved through extended rallies and deep corrections in a roughly four-year rhythm. Armstrong said he believes that pattern still holds, that the current cycle has likely bottomed, and that prices could rise next year.
He also argued that the price of bitcoin is no longer the only measure of how the industry is doing. He pointed to growth in stablecoins, tokenized stocks and blockchain-based financial services as evidence that adoption is expanding even while bitcoin lags. That view fits the direction Coinbase itself has taken, with a growing share of fee revenue coming from prediction markets and derivatives and a new bet on payments between machines.
