You'll buy your suppliers' agents before you buy their software

Last quarter you bought a piece of software from a supplier. You sat through the demo, argued the seat count, signed an order form, and waited for the login to arrive. Then your team spent six weeks learning where the buttons were. The thing you paid for was a place to do the work yourself, faster.

The next thing you buy from that same supplier won’t have buttons. It will have a counterpart. Their agent talks to your agent, the two of them reconcile the order, the delivery date, the disputed line item, and the first time a human looks, the work is already settled.

The next vendor relationship is their agent and your agent settling work between themselves, and procurement has no playbook for it.

What you actually buy from a supplier today

Set aside the product for a second and look at the relationship. What flows between you and a supplier is not really software. It’s a stream of small reconciliations.

They send an invoice; someone on your side checks it against the PO. They quote a lead time; someone follows up when it slips. A shipment arrives short; someone files the discrepancy, someone on their side confirms it, a credit memo eventually appears. Multiply that by every supplier you have, and the real product you’re buying is a steady supply of humans on both ends keeping two systems in agreement.

The software was always incidental. You didn’t want a portal. You wanted the order to be right.

You never wanted the supplier’s software. You wanted the work between you settled correctly, and you were willing to log in to make it happen.

That’s the part worth naming before anything changes: the value in a supplier relationship lives in the reconciliations, not the login. Hold onto that, because it’s about to move.

The naive version: bolt a chatbot onto the portal

Here’s the upgrade every vendor is reaching for first, and it’s worth watching it fail before we go anywhere else.

A supplier adds a chat box to their portal. Now instead of clicking through six screens to check an order status, you type “where’s PO 4471?” and it answers. Genuinely nicer. The demo lands, the buyer nods, the box ships.

Then the real work shows up, and the box just sits there. The invoice arrives with a quantity that doesn’t match what was delivered. The chatbot doesn’t notice, nobody asked it. It’s still waiting, politely, for someone on your side to open the portal, spot the mismatch, type the question, read the answer, and then go do something about it in a different system entirely. The supplier automated the part where you ask a question. They left untouched the part that was actually expensive: someone noticing the discrepancy and chasing it across two companies’ systems until it’s resolved.

The chatbot lives inside the supplier’s walls. It can answer about their world. It can’t see your world at all, your PO, your receiving log, your tolerance for a late shipment this week versus last. So the reconciliation still falls to a human, and that human still does the cross-company legwork by hand. A faster question is not a settled order.

The naive fix made the interface better and left the relationship exactly as labor-intensive as before. The work between the two companies still needs a person to carry it across the gap.

The shift: you buy a counterpart, not a portal

Now change one thing. Don’t buy access to the supplier’s software. Buy an agent that acts on the supplier’s behalf, and let it talk to an agent that acts on yours.

The supplier’s agent knows their world: inventory, lead times, what they’re allowed to credit without escalation, which late shipments they’re contractually on the hook for. Your agent knows your world: the open POs, the receiving logs, this month’s cash position, which orders are urgent because a customer downstream is waiting. Neither agent sees inside the other’s company. They meet in the middle, the way two people from two firms meet on a call, except they’re awake at 3am and they never forget a single open line item.

The invoice arrives with a mismatched quantity. Your agent notices, because watching the receiving log against incoming invoices is exactly the vigilance it’s built for. It raises the discrepancy to the supplier’s agent. The supplier’s agent checks the shipment record, confirms three units were short, and proposes a credit it’s pre-authorized to issue. The two agents agree, draft the credit memo, and post the correction to both systems. When a human on either side looks in the morning, the line item isn’t a task in a queue. It’s a resolved entry with a clean trail of how it got resolved.

The key idea is simple. The product stops being a place you log in and becomes a counterpart you negotiate with. You’re not buying the supplier’s software anymore. You’re buying their agent, and the unit of value is no longer “access to their system” but “work settled with their system, without a human carrying it across.”

This is why we think the order form changes before the software does. You’ll be paying for a counterpart long before anyone redesigns the portal you used to log into.

What procurement has no playbook for

Here’s where it gets uncomfortable, and where the honest version of this post earns its keep. Buying a counterpart is not buying a tool, and almost nothing in how companies buy software was built for it.

When you buy a tool, the questions are familiar. What does it cost per seat. What’s the uptime. Who owns the data. Can we export it if we leave. Procurement has decades of muscle for those. None of them answer the questions a counterpart raises.

What is the supplier’s agent allowed to commit to on their behalf, and how do you verify that boundary, not just trust the marketing? When your agent and theirs disagree, whose record wins, and what’s the appeal? If their agent agrees to a credit and a human later disputes it, is that agreement binding, did two pieces of software just sign something? When the two agents settle a thousand line items a month with no human in the loop, who’s accountable when the thousand-and-first is wrong in a way that costs real money?

Suppose a mid-sized accounting practice wires its agent to a payroll supplier’s agent so the two can reconcile filings automatically. Within a month they’re settling a few hundred line items between them, untouched by a human, exactly the win everyone wanted. Then one settlement adjusts a withholding figure that turns out to be wrong, and the question lands cold on someone’s desk: who agreed to this? Not “which person.” Two agents agreed, each acting inside a scope nobody had written down precisely, and now a regulator-facing number is off. That’s not a reason to stop. It’s the reason the boundary has to be explicit before the first settlement, not reconstructed after the bad one.

These aren’t reasons not to do it. They’re the actual work of doing it well, and a chat box lets everyone pretend they don’t exist.

Procurement knows how to buy a tool and how to hire a person. A counterpart is neither, and that’s the gap nobody has a contract template for yet.

The naive answer is to treat the agent like software, same MSA, same SLA, same security questionnaire, and discover too late that an SLA says nothing about what an autonomous counterpart is permitted to agree to. The other naive answer is to treat it like a person and try to supervise every move, which throws away the entire reason you bought a counterpart in the first place.

The real answer is the one that’s hard to put in a template: earned authority with a clear boundary and a clean record. The supplier’s agent gets a scope, these line items, up to this amount, within these terms, exactly the way a salesperson has signing authority up to a limit. Above the line, it escalates to a human on their side. Your agent has the mirror-image scope on yours. Every settlement between them leaves a trail you can read after the fact, so “the agents agreed” is never a black box, it’s a record of who proposed what, under what authority, and where a human could have stepped in.

Get that right and the relationship gets more trustworthy, not less, because for the first time there’s a record of every small agreement, instead of a credit memo someone vaguely remembers approving over the phone.

The turn: the relationship was always between people

There’s a quiet fear under all of this, and it’s worth saying plainly. If your agent and your supplier’s agent settle the work between themselves, what’s left for the humans?

Everything that mattered in the first place.

Two companies don’t have a relationship because their invoices reconcile. They have a relationship because someone decided this supplier was worth betting on, that their quality holds under pressure, that they’ll pick up the phone when a shipment is on fire, that the terms are fair enough to build on for years. That judgment was never in the line items. It was in the line items being a tax that ate the time you’d rather have spent deciding whether this is a partner worth deepening at all.

Move the reconciliations off the humans and onto two scoped agents, and what’s left on your desk isn’t less. It’s the part that was always the actual relationship: who to bet on, when to walk away, which supplier to bring closer because they’ve earned it. The agents settle the thousand small agreements. You decide whether there should be a relationship at all, and that has never been a thing you could buy a seat for.

The buttons were never the point. The buttons were the toll you paid to get to the work, and the work was never the reconciliation either. It was the trust the reconciliation kept getting in the way of.

---

That’s what we’re building at Apollo Space: a company that runs on agents, so the counterpart on the other end of your next supplier deal can be one too, settling the work between you, leaving a record you can read, and handing back the only decision that was ever yours. The order will already be right when you wake up. What’s left for you is choosing who’s worth ordering from in the first place.