The last letter closed on a decision you will make over the next twelve months whether you mean to or not: which agents become part of how your business runs, and which ones you let pass. This letter is about the shape of that decision. The technologies operators rely on tend to arrive as point solutions, and over time they consolidate into integrated platforms. Agents will follow the same arc. The CEO question is where on that arc to commit, and what trade-offs come with each position.
This is not an argument that platforms are the right answer. There are good reasons an operator might decide otherwise, and they are set out below in the same terms as the case for the other direction. The work is to see the arc clearly enough to choose deliberately, rather than to be carried along it.
The pattern, in your own business
The clearest example of this pattern in the operator’s world is the club management system.
What is now a single system of record was at one time four or five separate platforms. Billing sat with a payments processor that knew nothing about your members. Access control was a turnstile vendor with its own software and its own user list. CRM was the membership team’s spreadsheet, then a generic CRM bolted on the side. Member management was wherever the front desk happened to record it. Each tool was the best version of itself when it arrived. Operators ran on stacks of them for years.
The thing that made consolidation inevitable was not, primarily, the operator’s frustration with running four systems. It was the commoditisation of the core. The primary focus of a club management system was payment processing, and processing differentiation came down to cost per charge. You cannot build a durable competitive position inside a single-digit basis-point spread. It became logical, and then necessary, for vendors to expand beyond that core. CRM, member management, access control. Each expansion was a move to differentiate, and at the same time a move to bring the disparate parts of an operator’s technical stack together under one platform. The operator’s integration pain was real, but it was downstream of the vendor’s commoditisation problem. The two converged.
Four forces then made the consolidation stick.
Integration cost outgrew tool value. Every new tool meant another connector, another login, another place for data to drift. The cost of running the stack grew faster than the value of any single tool inside it.
The data fragmented. The member who churned was a different member in the booking tool, the billing tool, the CRM, and the access system. Decisions made from any single view were decisions made on a partial picture.
Governance stopped being coherent. Permissions, audit trails, accountability. Each new tool brought its own version of those questions, and multiplied across the stack the answers no longer hung together.
The ownership question reached the CEO. The executive meeting conversation turned into a pattern any operator will recognise. Why is this so hard. Why do the pieces not work together. Why are we paying third parties to integrate our own stack. Why are we paying our vendors extra to talk to each other. Who in this organisation actually owns the outcome when a process runs across four systems? Those questions are the moment the conversation changes. They are not about cost. They are about accountability, and the absence of an answer is what makes them land.
The first move a club management system makes after it acquires another application, or another company with a parallel application, is to announce a new platform. The announcement often runs ahead of the work, and what gets shipped is a loosely-fitted set of jigsaw pieces rather than a true platform rebuild. The pattern was visible from the outside. What was happening underneath it was the commoditisation pressure that made the move necessary in the first place.
Operators who recognised the arc early made better calls. Operators who recognised it too late paid twice: once for the point tools, and then again for the integration work, the migration cost, the re-training of teams on the new substrate, and the price of having committed late to a vendor whose pricing power had risen by the time they signed.
The same mechanic played out in the wider category. Standalone customer databases became CRM platforms. Standalone marketing tools were folded inside them. Support tools followed the same trajectory inside service platforms. Reporting moved from a person with spreadsheets to an analytics layer sitting across the stack. Each time, the point solution survived only as long as its value outran the integration cost, the data fragmentation, the governance gap, and the customer’s question of who was accountable for the whole.
The same arc, applied to agents
The early shape of the agent market today is likely point solutions. I expect there will be vendors building specialist agents covering areas such as sales, retention, collections, and the other categories of work that sit inside an operator’s business. There will also be vendors who have built from a different starting point, treating the platform shape as the right place to begin and the agents as activations on top of it. Both shapes are in the market now. The operator’s decision is not whether to wait for the arc to play out. The arc is already underway, and the question is which shape to commit to first, knowing what each one is built to do well.
Agents will follow the same arc as the categories before them, and the four forces from earlier waves will act on this one too. The data still fragments when each agent holds its own view of the member, the lead, the coach, the schedule. The governance question lands harder than before, because an agent that books a class, refunds a payment, or changes a member’s status is making a decision in your name. The ownership question reaches the CEO sooner, because the same executive-meeting pattern returns with agents in place of tools.
The integration mechanics, in this wave, will get easier rather than harder. Every platform of this kind being built now is built API-first and designed to be called by other systems and to call them in return. The legacy integration pain operators carry came from a generation of tools that were not built that way. It is not coming back. The cost is not the wiring. The cost is what does not happen when agents only call each other rather than learn from each other.
Two further forces sit on top of those, specific to agents, and they are the ones the analogy with previous waves understates.
The knowledge base multiplies with every point agent. Every agent that does useful work in your business needs to understand what your business is, how it operates, what its products are, what its tone of voice is, what its policies are, and which questions are escalated rather than answered. That is the agent’s knowledge base, and it has to be built and maintained. With one agent, you build it once. With four, you build the same body of knowledge four times and maintain four copies as the business changes underneath them. Previous waves did not have this problem because the tools were configured, not informed. Agents are the first category that needs to be told what the business is in order to act on its behalf.
The lifecycle of the member runs across agents. A member moves through your business in a sequence: arrival through sales, the membership relationship itself, sometimes collections, then a retention conversation, and regrettably sometimes cancellation. Each of those stages is a credible point-agent category. Acting in isolation, the agents make sensible decisions inside their slice. Acting together, with a shared view of the member’s history across every stage, they produce a different kind of outcome. The retention agent that knows what was promised at sign-up has a different conversation. The cancellation agent that knows collections resolved a payment dispute two weeks ago has different timing. The sales agent that knows this prospect was saved by retention at fourteen months last year, then chose to come back twelve months after leaving, has a meaningfully different conversation when they walk in. One plus one stops equalling two and starts equalling three. That compounding is the value the lifecycle produces when the agents share what they learn. It is the value left on the table when they do not.
Agents are also unusually portable. They can be moved between systems more easily than the tools they replace could, because the heavy lifting is done by general-purpose models that already understand language, intent, and context. That portability could keep the point-solution phase going for longer than it did for previous categories. The arc still bends in the same direction. It may simply take a little longer to bend.
The case for point agents
There is a real case for buying best-of-breed point agents, and no benefit to anyone in pretending otherwise. The arguments below are the right arguments when each agent is judged on its own job, in isolation from the others.
Speed of value. A specialist agent built for a specific job by a vendor that does only that job may be better at that job sooner than a platform offering the same capability as one of many. The word may matters. The speed of value the vendor delivers on the single job is not always the speed of value the business actually feels, because what the business feels is the lift across the lifecycle of the member, not the speed inside a single slice of it.
Optionality and pricing. Point agents are easier to swap, and they keep costs proportional to the agents you actually rely on. Two things to hold alongside that. A vendor whose customer relationship is shallow by design has weaker incentive to invest in the long-run depth of the agent the customer bought. The platform that is harder to leave is, for the same reason, more committed to making it worth staying. And lock-in itself is a commercial-model question, not a structural property of platforms. Some platforms are sold on per-seat licences that commit you to functionality you may not use. Others are sold on consumption or outcome-based models that charge you only for the work actually performed. A future letter goes deeper into the shape of commercial models for agentic work, because the move from licence pricing to consumption and outcome pricing is, in its own right, one of the more consequential shifts of this wave.
Specialisation per function. Specialisation usually beats generalisation, but the question is specialisation in what. A cancellation agent specialised in cancellation conversations, with no view of what the sales agent promised, what the collections agent resolved, or what retention already tried, is specialised in the wrong unit. It is specialised in the conversation, not in the outcome. A platform-native agent that is slightly less specialised in the conversation but is specialised in producing the right outcome for this specific member’s lifecycle is the better tool for the job the CEO actually cares about.
These are not weak arguments. They are the reason point-tool phases of every previous wave lasted as long as they did. The question is becoming whether the agent-by-agent lens is the right one to judge by at all. A best-of-breed point agent judged on its own job will often win that comparison. The same agent judged on the lift it produces across a business running several agents will often lose to a competent platform-native agent that is compounding with the rest.
The case for a platform
The case for a platform is not a case for easier integration. Integration mechanics are going to get easier for everyone. The case for a platform is the case for what happens after the pieces connect. Agents running on a shared substrate observe each other, learn from each other, and lift each other’s performance through that shared context. Agents wired together do not.
The highest-performing operator businesses you have ever worked inside share something specific. They run on one record of truth. Everyone is on the same page because they have full transparency of what is happening in the business in real time, and the opportunity to learn from the experience of every other function across the lifecycle of the customer. A decision made in the retention conversation is informed by what happened at sign-up, what happened at the front desk last week, and what the collections process resolved last month. The lowest-performing businesses are the opposite. Each function holds its own truth and acts confidently from inside it. The first organisation outperforms the second by a wide margin, and the difference is not the quality of the people in each function. It is the substrate they share.
The same difference separates a business running point agents wired together from one running agents on a shared platform. A platform holds the member’s history across every stage as one continuous record. Each agent acts with the context every other agent has produced. The agents observe each other and lift each other’s performance in real time. The knowledge of what the business is, how it operates, and how it speaks lives in one place, is maintained in one place, and is referenced by every agent. Permissions, audit trails, escalation rules, and the boundary between what an agent may decide and what must come to a human are configured once and apply across the whole. None of this is an argument for a particular platform. It is an argument for the platform shape of solution, once the work running on it has crossed the threshold where the forces above begin to bite.
Where on the arc to commit
The decision the operator can actually picture today is the first one. The reality is that almost nobody is running three, four, or five agents in their business right now. Most are running one, or are about to.
Two tests for that first decision.
The first test is the obvious one. Evaluate the agent on its ability to deliver the job you are buying it for. That is the test most operators will run by default, and it is the right starting test.
The second test is the one most operators will not yet think to run. Price into your decision the picture of what your business looks like when this is no longer one agent but three or four. Every operator business will, within a small number of years, be running multiple agents across multiple functions. The question is not whether that future arrives. It is whether the agent you are about to commit to is built to thrive inside it, or only built to thrive on its own. The first test tells you whether the agent is good at the job today. The second test tells you whether the agent will still be the right purchase once the rest of the business has agents in it too.
Operators who run only the first test will buy well for this quarter and find themselves in the consolidation conversation sooner than they expected. Operators who run both tests will buy with the arc in mind, and arrive at the multi-agent business already shaped for it.
