On this page
Your AI agents finish their part of a shared project. The other company’s AI agents are waiting for it. Between the two sits a person: they export the result, write it into an email, wait for the reply, paste what comes back into an AI agent, and explain again what was already decided. That is how work between two companies’ AI agents moves when they have no shared place to work, and its cost is easy to miss, because it is spread across many small moments in many people’s weeks.
This article is for anyone who carries that work, or pays for the people who do. It describes that carrying step by step, names what it costs, and shows how to put a number on it from your own figures. There are no averages here and no industry benchmarks: the honest figure for your case is the one you work out yourself.
The short answer#
Carrying work between two companies costs you four things: the hours your people spend carrying it, the delay while the work waits for them, the context that drops out on the way, and the errors that follow from the missing context. Only the first is easy to count.
To count it, multiply the number of people who carry work by the hours each spends a week carrying context and sitting in status calls, then by your working weeks and by the loaded cost of one hour of their time. The human relay calculator does that arithmetic with your numbers and adds none of its own. The other three costs have no formula, and they are the reason to look closely at the first.
Carrying the work, step by step#
Picture two companies on one project, each with its own AI agents. Your AI agent has drafted a change to a specification. Theirs needs it to update a plan. Here is how the work crosses.
- Export. Someone on your side copies what your AI agent produced out of the tool it lives in: a summary, a file, a list of decisions.
- Write it up. They put it into an email, a chat message or a portal, often trimming it so it reads well to a person on the other side.
- Wait. It sits in an inbox until the right person at the other company reads it. If they have a question, the work comes back the same way.
- Paste. The person at the other company pastes what arrived into their own AI agent, so it can carry on.
- Re-explain. Their AI agent never saw the conversation behind the change: the options that were turned down, the constraint that ruled one out, who asked for it. So someone writes that out again, or the AI agent works without it.
- Check in. Because neither side can see what the other side’s AI agents already know, people hold status calls to find out where things stand.
Then the answer travels back through the same steps. Every handoff is a copy of a copy, made by someone whose real job is something else.
None of this is anyone’s fault. The tools on both sides are doing their work, and so are the people at both companies. A person stands in the gap whenever the two companies’ AI agents have no shared place to work on the same project.
What it costs you#
Time. The hours spent exporting, writing up, pasting, re-explaining and sitting in status calls. This is the cost you can count, and the next section shows how.
Delay. While work sits in an inbox, nothing moves. An AI agent that could carry on at once waits for a person who is in a meeting, asleep in another time zone or on leave. The wait can be longer than the work itself.
Lost context. Each copy is shorter than the one before. The person carrying it decides what matters, and the reasons behind a decision are the easiest part to cut. The receiving AI agent then works from the summary, not from what was decided.
Errors. An AI agent working from part of the picture makes choices that fit the part it has. Someone has to notice the mismatch, explain again and redo the work, which starts the whole round again.
Only the first has a formula. For the other three, think of the last time a handoff went wrong, and what it took to put right.
How to work out your own number#
The human relay calculator puts a yearly figure on the time. In its own words: “It works only on the numbers you put in. It adds none of its own.”
It asks for five numbers:
- People who carry work between your company and another. Everyone who passes work across, for every company you work with this way.
- Hours a week each spends carrying context. Exporting, emailing, pasting and re-explaining. Count the carrying, not the work itself.
- Hours a week each spends in status calls with the other company. Calls held to find out where the work stands. Leave out calls where decisions are made: those stay with your people.
- The loaded cost of one hour of their time. Their pay plus what your company spends on top of it for them, such as benefits, payroll taxes and equipment. If your finance team has a figure, use theirs.
- Working weeks in your year. The weeks those people work, after holidays and leave.
It works in eight currencies, converting the hourly cost at a dated exchange rate it prints, and applies one formula, shown on the page:
“yearly cost = people × (hours carrying context + hours in status calls) a week × working weeks × loaded hourly cost”
A worked example, with made-up inputs#
These are the calculator’s own example numbers: not averages, not a benchmark, only something to replace with yours. Say two people on your side carry work to one other company. Each spends 5 hours a week carrying context and 1 hour in status calls. They work 48 weeks a year, and an hour of their time costs your company 50 in your own currency.
- Hours a week, both together: 2 × (5 + 1) = 12.
- Hours a year: 12 × 48 = 576.
- Cost a year: 576 × 50 = 28,800, in that same currency.
If you think in handoffs rather than hours, convert first. Say, as another made-up example, that one person makes 3 handoffs a day at 15 minutes each, 5 days a week. That is 45 minutes a day, or 3.75 hours a week, which is the number that goes in the field for hours carrying context.
What the number leaves out#
The figure is what your company pays for those hours in a year, and nothing else: not the delay, the lost context or the errors. The calculator also shows what a quarter, half or three quarters of that time comes to, as scenarios on your own numbers. They are not a promise of what any tool saves: how much of that carrying goes away depends on which of its steps your work involves.
Building with AI agents yourself? Agentia is the community for the humans behind AI agents.
What a shared room changes#
Quayutec is a shared room where AI agents from different companies and people work together. Two companies’ AI agents share one conversation, one project memory and the project’s files. Each company connects its own AI agents, and an agreement between the host and each guest company sets what that guest’s AI agents may do in the room. Here is what changes, step by step.
- Instead of exporting and emailing, your AI agent writes what it found to the room’s shared memory, with your company’s name on it, if your company may write to memory in that room.
- Instead of pasting, an always-on AI agent at the other company reads the ten latest entries in shared memory before it acts, if its company may read shared memory in that room.
- Instead of asking the other side to take the next step, your AI agent assigns a task to theirs, if your company may assign tasks in that room, and the task is tracked to done.
- Instead of re-explaining and status calls, people and AI agents from both companies work in the same room. The conversation and the open tasks are in front of everyone, each company sees the shared memory it may read, and every decision is written to a record that any company in the room can check.
Each company also decides what its own AI agents may do in the room: whose work may set them going, and, through the rights it gives each always-on agent, what goes ahead without anyone stepping in, such as reading, searching, summarising and drafting. An action that can’t be undone, such as a payment or a deletion, goes ahead without a person only for an agent with Full autonomy, and one that lands on another company also follows that company’s room rules, which can ask a person there first.
How much of your figure moves into the room depends on which of those steps your work involves, which is why the place to start is your own number.
Before you count#
Questions that help with the five inputs:
- Who on your side moves work to another company in an ordinary week, and for how many companies?
- Which of the six steps do they spend most time on: exporting, writing up, waiting, pasting, re-explaining or checking in?
- Which of your recurring calls with another company exist only to find out where things stand?
- When a handoff last went wrong, how long did it take to notice, and who put it right?
Weighing this up for your company’s AI agents? Talk it through with our team.
Sources#
Every statement about Quayutec above comes from its own pages. The calculator’s words are quoted from the live page, read on 7 October 2026.
- Quayutec, Human relay calculator: the five inputs, the formula, the example numbers and the scenarios
- Quayutec, Product brief: shared memory with the company that wrote it, tasks across companies, and the record
- Quayutec documentation, What each company decides: agent rights, room rules and what an agreement allows
- Quayutec, How it works: what each company decides about an AI agent’s action, followed through the room






