We believe agents will run the repeat work of every team.
What we believe.
Every company has work that comes back every week: finding the information, sorting it, preparing the next step, chasing the reply, remembering what was said, and reporting on it. We believe agents will do that work, inside the software a team already uses. Your people keep the decisions, the relationships and the creative calls.
A company gets there when someone puts the agents into its work and keeps them current, not when each person is handed another chatbot. That is the part we do, and the rest of this page explains why we think it, what it looks like in a team’s week, and where we differ from others who write about it.
About six minutes. Every outside quote is word for word, dated and linked.
What changes in a company’s week.
Today, each person in your company probably has a chatbot beside their work. It helps that person with that task, and what it learns stays with them. Ten people have ten chatbots, and nothing connects what any of them knows.
The next stage puts the agents on the work itself. The work of most teams runs in the same six steps, over and over:
- FindPull what is needed from the places it lives.
- SortDecide what it is and where it goes.
- PrepareDraft the next step so it is ready to review.
- ChaseFollow up until the reply arrives.
- RememberKeep what was learned where the whole team can use it.
- ReportSay what happened, to whoever needs to know.
People keep
- Decide
- Relate
- Create
Agents take on those six. People keep the row beneath them: Decide, Relate and Create. Finished work waits for review, and the decision is always a person’s.
The effect shows up when the company takes on more. A new team, a new client or a new line of work starts with agents already working, and what was learned is already in the company’s memory. The same people carry more of the work, so growth does not wait on hiring to match.
Example, labelled as one. In an agency, the loop runs once per client. That is why an agency’s hiring has followed its client list, and why it can follow it less closely once the repeat work is carried by agents.
Why now.
Building and changing a system takes days. It used to take months. A system fitted to one company’s work can now be built, tested with the people who use it, and changed when the work changes, in the time a team used to spend waiting for a specification. That makes it reasonable to build a system for one company, around how that company already works.
The frontier keeps moving. New models arrive often, and some of them do a company’s work better than the one it is running. A system built once falls behind, quietly, while the people who depend on it are busy with other things. We think that is the larger risk in this field: the system works on launch day and nobody owns what happens afterwards.
This is why we think of the work as a partnership that continues. Someone has to watch the frontier, pick what is ready, test it on the company’s own work, and move the system only when the result is better. Your team should not have to track every release for that to happen.
How trust is earned.
A company should not hand work to an agent on a promise. We think an agent earns its freedom one step at a time, for each stage of the work, on evidence from that company’s own work.
- WatchThe agent shows you what it sees. You do the work.
- SuggestThe agent proposes the next step. You decide.
- Act within limitsThe agent acts, inside the rules you set.
- RunThe agent runs the stage. You review the results.
Each move up is made on evidence, starting with the stages that are easiest to reverse. At every step you can move an agent back down, and the work it did can be checked. Questions such as “what would you let a system do without asking you?” have different answers in different parts of the same company, and the ladder lets each part answer for itself.
What we do about it.
As your implementation partner, we work inside your team. We learn the work where it happens, build the system around it, and connect it across the teams that touch the same work. The value of agents is in that fit: a product bought for each job is another seat and another login, and none of them knows how your company runs.
As your operating partner, we stay. We run the system, measure it against the baseline we set at the start, and keep it at the frontier: when a newer model does your work better, we move the system onto it, and when it does not, nothing changes. Your people keep the decisions, and our team keeps the technology current.
That is how a company becomes AI-native: through agents on its repeat work, built into its software, run by someone who stays. We help your company leverage agentic technology without asking it to follow every release.
OCTYN runs its own work on its own systems. We work with the same loop and the same ladder that we describe here, which is how we know what keeping a system current takes.
The outside view.
Investment firms and AI developers have been writing about this since 2024. Here is what they wrote, word for word, in date order. Each quote links to the page it came from, and each is followed by our reading, in our own words, kept apart from theirs. The firms have no relationship with OCTYN. Quotations are from public writing, last checked 16 October 2026.
“In the services business, responsibility for achieving the desired outcome sits with the company selling the service.”
Our reading: It is why we stay and run what we build.
“AI led services - a strong technology product with humans in the loop - will dramatically change how services are delivered.”
Our reading: People stay in the loop, which is where we put them.
“How you integrate, embed, and operate becomes the moat.”
“The traditional ‘deploy and maintain’ model has become ‘deploy and continuously re-engineer.’”
Our reading: This is the case for an operating partner. The work after launch is where the value is kept.
“The AI applications of 2026 and 2027 will be doers. They will feel like colleagues.”
Our reading: Agents on the work, beside the people who review it.
“A copilot sells the tool. An autopilot sells the work.”
“AI has crossed the threshold where it can do most of the intelligence work autonomously and leave the judgement to humans.”
Our reading: The first line is the one on our homepage. The second says what we mean by “people keep the decisions”.
“Companies from community banks to mid-sized manufacturers and regional health systems stand to gain from AI, but lack the in-house resources to build and run frontier deployments.”
Our reading: Access to the models is not what is missing. Putting them into the work, and running them, is. Anthropic and the firms it works with serve the same kind of buyer as we do, so we cite them as one view among several.
Sequoia Capital is among the backers named in the Anthropic announcement, and Foundation Capital invests in companies of the kind it describes. These are positions, and we show them so you can weigh the writing accordingly. Quotations are from public writing. Their authors have no relationship with OCTYN.
Where views differ, and where we stand.
The people writing about this do not agree on everything. We think that is useful to know, and it shows where our own position is a choice.
Buying companies or working inside them
Some investment firms back companies that buy existing service companies and put agents into them. Elad Gil described this in an interview with TechCrunch on 1 June 2025 (read it →). Sequoia, in the March 2026 piece, writes about firms that take over an outsourced task: “The outsourced task is the wedge.”
Where we stand: We own no one’s company and take no one’s work away. We work inside the company that has the work, and the work stays there with its people.
Pricing by outcome
Foundation Capital expects pricing to move toward outcomes, and says: “The progression toward outcome-based pricing will be messy, with false starts, hybrid tiers, and plenty of margin surprises.” (3 July 2025, link)
Where we stand: We agree it is messy, and we measure each system against a baseline set with you at the start, so you can see what it does. We publish no promise that a number will move.
What the margins look like
Some of the writing projects software-like margins for AI-led services. Lightspeed, an enthusiast, is cautious: “saying ChatGPT can code so IT services margins should be 90% is too simplistic.” (18 November 2024, link)
Where we stand: We make no margin claims. Where a system has run, we show exact figures from it as receipts, and share the full receipts on request.
Where the line between machine work and people’s work sits
Sequoia expects the line to move: “Today’s judgement will become tomorrow’s intelligence.” (5 March 2026, link) Lightspeed expects people to be needed for a long time: “we believe that while we get closer to automation a large number of services will require human vetting or augmentation for the foreseeable future.” (18 November 2024, link)
Where we stand: Both can be true, so we do not fix the line. The trust ladder moves it one stage at a time, on evidence, and the person who owns the work decides when.
What we do not believe.
That people get replaced. The work loop ends in a row that only people do: deciding, relating, creating. Agents take the repeat work so that those hours go further, and review stays with a person at every step.
That becoming AI-native is a threat to be faced. It is an outcome a team can choose, at its own pace, one stage of its work at a time: the agents are built in, run by someone who stays, and kept current. Each stage can be stepped back from, so starting commits a team to very little.
Questions.
- What does OCTYN believe about where work is going?
- We believe agents will run the repeat work of every team: finding, sorting, preparing, chasing, remembering and reporting. People keep the decisions, the relationships and the creative calls. A company gets there when someone builds the agents into its work and keeps them current.
- Will agents replace our people?
- No. The work loop ends in a row that only people do: deciding, relating and creating. Agents take the repeat work, and finished work waits for a person’s review.
- Why not give everyone a chatbot, or buy a product for each job?
- A chatbot helps one person with one task, and what it learns stays with them. A product for each job adds another seat and another login, and none of them knows how your company runs. Agents on the work are fitted to your processes, connected across teams, and kept running by a partner.
- What is an agent allowed to do without asking?
- Only what you have let it do, stage by stage. An agent starts by watching, then suggests, then acts inside limits you set, then runs while you review. It moves up on evidence from your own work, and you can step it back at any point.
- How do you keep a system current as models change?
- We measure each system against a baseline set at the start and re-set it as your company changes. When a newer model does your work better, we move the system onto it. When it does not, nothing changes.
- Are the firms you quote connected to OCTYN?
- No. Sequoia Capital, Foundation Capital, Lightspeed Venture Partners and Anthropic have no relationship with OCTYN. The quotations come from their public writing, linked on this page so you can read them in full.
See what agents would take on in your team.
On a mapping call we look at one team’s week, name the repeat work, and say which of it an agent can carry today and which should stay with your people.