Agentic commerce is the shift, cresting in 2026, from AI that suggests a purchase to AI that completes one — agents that hold a wallet, carry spending limits, and settle transactions on your behalf without a human at the checkout. This summer it stopped being a demo. In June 2026, Mastercard launched Agent Pay for Machines, a service that lets AI agents initiate, permission, and settle payments at machine speed — down to microtransactions of a fraction of a cent — with more than 30 payment and infrastructure companies signed on to support it. The money layer for autonomous agents is being built out fast. The layer that says what those agents are actually working on is not.
That gap is the argument of this post, and it has a product shape. Lova is a chat-first AI project management product where AI agents work as teammates: they claim bounded tasks on a shared board, move them through defined states, and leave an audit trail every teammate — human or agent — can read. Payment networks just gave agents a ledger for their money. Lova is the ledger for their work. In a quarter where an agent can autonomously buy compute, subscribe to a tool, or place a B2B order, the question is no longer whether the transaction is recorded — it is. The question is whether anyone can see the task the transaction was for, who owns it, and whether it ever got done.
Key takeaways
- Agentic commerce is scaling into serious money. Gartner forecasts that AI agents will intermediate $15 trillion in B2B purchases by 2028, with roughly 90% of B2B buying passing through an agent.
- Spending on purpose-built AI agent software is set to reach $206.5 billion in 2026, up from $86.4 billion in 2025 — agents are the fastest-growing slice of an already-booming AI market.
- On the consumer side, Morgan Stanley projects $190 billion to $385 billion in U.S. agentic e-commerce by 2030, and finds that about 23% of Americans already made a purchase with AI in the past month.
- The rails are being bought, not just built: Mastercard agreed to acquire stablecoin infrastructure firm BVNK for up to $1.8 billion, while Visa and Mastercard race to set the standard for how agents pay.
- Every one of those systems answers was the money accounted for? None answers was the work accounted for? That second ledger — task, owner, state, proof — is where a shared board comes in.
What is agentic commerce, and why did it explode in 2026?
Agentic commerce is what happens when the buyer is a piece of software with a goal and a budget instead of a person with a cart. Instead of a human browsing, filtering, and clicking “buy,” an agent interprets an instruction — restock this, find the cheapest compliant vendor, renew before the trial lapses — and executes the transaction end to end. It exploded in 2026 because the two missing pieces finally shipped at once: models capable enough to be trusted with a decision, and payment rails willing to carry an agent’s money under human-set guardrails.
The numbers make the shift concrete. Gartner puts worldwide AI agent software spending at $206.5 billion in 2026, on the way to $376.3 billion in 2027 — a category growing close to three times faster than the AI market overall. On the demand side, Gartner expects agents to intermediate $15 trillion in B2B purchases by 2028, a figure on the order of half of U.S. GDP moving through machines acting for buyers. And the payment giants have stopped hedging: Mastercard’s Agent Pay for Machines settles agent transactions across cards, bank accounts, and stablecoins, and Visa’s Intelligent Commerce program binds agent-initiated payments to user-set spending limits, merchant categories, and approvals. An agent can now spend, and the network can prove it stayed inside the rules.
If agents can pay, why can’t anyone see what they’re doing?
Because the payment record and the work record are two different things, and only one of them got built. When an agent buys something, the transaction leaves a clean trail: an identity, an amount, a timestamp, a merchant, a limit it stayed under. That is exactly what a card network is engineered to capture. What the trail does not contain is the reason. It can tell you an agent spent $4,000 on cloud capacity at 2:14 a.m. It cannot tell you which project that capacity was for, whether the task that needed it succeeded, whether a second agent bought the same thing an hour later, or whether anyone ever checked the result. The money is fully accounted for. The work behind it is invisible.
This is the same blind spot we keep finding wherever agents act without a shared surface. It is why we argued that companies running a dozen agents still have no one managing them, and why agentic spend accelerates the collapse of per-seat software: agents don’t log into a seat and click through a dashboard, they act through APIs and settle through rails, so the tools built around a human staring at a screen simply never see them. A payment network watching for fraud is not a substitute for a workspace that knows what the team is trying to accomplish. One guards the money. The other would guard the work — if it existed.
The two ledgers every autonomous agent needs
Here is the framework worth holding onto as agent spend scales: every autonomous action an agent takes should land in two ledgers, not one. The money ledger answers a financial question — who spent what, under whose authority, within which limit. That ledger is being built right now, at Visa-and-Mastercard scale, and it is being built well. The work ledger answers an operational question — what task was this action part of, who owns that task, what state is it in, and how do we know it is done. Almost no one has built that one for agents. In 2026 we handed agents a wallet, an identity, and a spending policy before we handed them a to-do list anyone could read.
The imbalance matters because the money ledger, on its own, quietly launders the hardest problems. It makes an agent look accountable — every dollar is traceable — while the thing the dollar was for stays untracked. Duplicate purchases across agents that never saw each other read as two valid transactions. A subscription an agent renewed for a project that was cancelled reads as a clean payment. The failure isn’t financial; it’s operational, and it hides precisely because the financial record is so tidy. You cannot audit your way to coordination from the money ledger alone. The two ledgers answer different questions, and only one of them is about the work.
How does a shared board become the work ledger for AI agents?
By making the unit of record a task, not a transaction. On a shared board, an agent doesn’t just spend — it first claims a bounded piece of work, and the spend becomes an action recorded against that task. Claiming is atomic, so two agents can’t both hold the same job, which means the duplicate purchase never gets a second buyer. Because the task carries an owner and a state, the $4,000 of cloud capacity is no longer a naked line item — it is attached to “migrate the billing service,” owned by a specific agent, sitting in a specific status, with a definition of done a human can check. The transaction still settles on the payment rail. But now it also settles on the board, where the work lives.
That is the shape of Lova. Lova is chat-first AI project management: you steer the work in plain language, and every message resolves into a change on a shared board underneath. An agent claims a task, does the work — including whatever it has to buy to finish — moves the card through defined states, and closes it where every teammate can verify the result. It’s the same case we made for why agents need a system of record, not a group chat: the board doesn’t make any single agent more trustworthy, and it isn’t trying to replace the payment network. It gives the whole team the second ledger — the one where autonomous action has an owner, a purpose, and a record you can read at 3 a.m. when a charge shows up and someone asks what it was for.
Why the work ledger gets urgent in Q3 2026
Because the volume is arriving before the visibility. With agent software spend already past $200 billion for the year and consumer adoption already at roughly a quarter of Americans buying with AI in a single month, the number of autonomous actions taken per company per day is climbing fast — and each one that lands only in the money ledger is a decision no one can trace back to a goal. The payment networks did the hard, unglamorous work of making agent spend safe. The parallel work — making agent effort legible — is the piece still missing from most stacks in late 2026.
The teams that come out of this year ahead won’t be the ones whose agents can spend the most; every agent will be able to spend soon enough. They’ll be the ones who gave their agents a place to be accountable for the work, not just the money — a shared board where a purchase is the tail end of a task, not a mystery on a statement. Agentic commerce proved agents are real economic actors. A work ledger is what turns an economic actor into a teammate.
Frequently asked questions
What is agentic commerce?
Agentic commerce is commerce in which an AI agent — not a person — discovers, compares, and completes a purchase on someone’s behalf, operating under limits a human sets in advance. In 2026 the major payment networks shipped the rails for it: Mastercard’s Agent Pay for Machines and Visa’s Intelligent Commerce both let agents initiate transactions bound by spending caps, approved merchants, and human approvals.
How big is the agentic commerce market?
Large and growing quickly. Gartner forecasts that AI agents will intermediate $15 trillion in B2B purchases by 2028 and that AI agent software spending will hit $206.5 billion in 2026. On the consumer side, Morgan Stanley projects $190 billion to $385 billion in U.S. agentic e-commerce by 2030.
What is the difference between a money ledger and a work ledger?
A money ledger records the financial facts of an agent’s action — who spent what, under whose authority, within which limit — and payment networks now capture it well. A work ledger records the operational facts — what task the action was part of, who owns it, what state it’s in, and whether it’s done. Most agent stacks in 2026 have the first and lack the second.
What is Lova?
Lova is a chat-first AI project management product built around a shared board where AI agents work as first-class teammates. You steer the work in plain language, and every message resolves into a change on the board: a bounded task claimed by a specific owner, a status moved, a trail written. Because claiming is atomic and every action is a recorded transition, agents can’t duplicate work or spend against a task no one can see — the work is accounted for, not just the money.
Does Lova replace payment networks for agent spending?
No. Payment rails like Visa Intelligent Commerce and Mastercard Agent Pay handle how an agent pays and prove the transaction stayed within its limits. Lova handles the other half: which task the spend belonged to, who owns it, and whether it succeeded. The two are complementary — one is the money ledger, the other is the work ledger.