Your CRM does not need more features. It needs an agent.

A CRM in 2026 is a finished product. Every serious vendor ships the same long list. Custom objects. Workflows. AI fields. Reports builder. Forecasting. Lead scoring. Calls. Emails. Tasks. Roll-ups. Permissions. A mobile app nobody opens. The feature pages on the four largest CRMs are now almost interchangeable when you read them side by side, which is what I do for a living.

Teams still spend hours every week doing work the CRM does not do for them.

That gap is not a missing feature. It is the entire shape of the problem.

This is the argument I have been making to founders for the last six months, and the one I think the next twelve months of CRM go-to-market will turn on. The interesting product in 2026 is not a CRM. It is the agent that sits on top of one.

The feature plateau is real

The mid-market CRMs hit a feature plateau around 2023. Salesforce, HubSpot, Pipedrive, Attio, Folk, Close. Each of them now ships every checkbox a buyer can find in a comparison matrix. Custom objects, workflow builders, reports, integrations, AI fields, MCP servers. The category has converged.

You can verify this by trying to write a comparison post. Five years ago, "HubSpot vs Pipedrive" wrote itself. Today the same comparison is mostly about ergonomics and pricing, because the feature lists overlap so heavily that the headline differences come down to who you call when something breaks.

The category is done growing. New features still ship, but each one is a small thing that almost no team will use to make a buying decision. The vendors know this. Their roadmaps in 2026 are mostly AI feature pages, because AI is the only place left where a feature can still feel like a step change.

Even those AI features run out of room fast. Once every CRM has a "summarize this call" button and an "AI field that classifies records," the comparison matrix re-flattens. The plateau extends by one row and the gap between products closes again.

The work the CRM does not do

Sit next to a sales team for a day and write down everything they do that the CRM did not do for them.

Most of it is not work that a new field would solve. It is work that requires reading something, deciding something, and writing several things in several places. A meeting just happened. The rep reads the transcript. They update three fields on the deal. They flag two follow-up tasks. They draft a recap email. They tag the deal as at-risk because the buyer mentioned a competitor. They post in Slack so the team knows the meeting moved the deal.

Six actions, all of them downstream of one judgment: "I just listened to this call and here is what changed."

That judgment is what the CRM does not have. It is also, until 2025, what no agent could reliably do either. The judgment requires reading the actual content of the call, holding the deal context in working memory, and writing back to the right fields in the right format. Each of those was a hard problem on its own and they had to be solved together.

In 2026 they are not hard anymore.

The model can read the transcript. It can hold the deal context. It can write to the right fields through the CRM's API or MCP server. The judgment-shaped work that used to require the human is now doable by an agent running on top of the same CRM the human was using.

The "missing feature" framing misses this entirely. The work the team is still doing is not a feature gap. It is a layer that did not exist yet and now does.

Three agents I have shipped that prove the point

I run Craftt. We are an AI-native Attio practice. The agents we ship are not theoretical. They are running on customer workspaces today, and each one closes a gap that no CRM feature is going to close, no matter how long the roadmap.

Pipeline Hygiene. Most CRMs let you build a list view called "deals at risk" with a filter. That tells you which deals to look at. It does not tell you what is wrong with them. Our Pipeline Hygiene agent reads every open deal, checks it against the team's stage criteria (which the CRM does not know), surfaces the specific reason each deal is stuck, and writes a one-line note to the deal record. Every morning. The list went from "thirty deals to triage" to "thirty deals, each with the actual reason," because the agent did the triage step that the CRM cannot.

Meeting Prep. Before a sales call, the rep wants the last five touchpoints with this account, the relevant signals from the prospect's company, and a recommended angle. The CRM has the touchpoints in the activity feed. The signals are in a Specter or Apollo enrichment. The angle requires reading both and saying something useful. The agent reads both and writes the angle into the deal record fifteen minutes before the meeting. The rep walks in prepared. No CRM feature would have produced that synthesis.

Content Mining. Every sales call holds a week of marketing content. Pain quotes, objections, vocabulary, before-and-after stories. The CRM stores the call transcript. It cannot extract those signals, because extracting them requires reading the call as a content marketer would, not as a sales rep would. The agent reads every transcript in the workspace, extracts five categories of seed material, and writes them to a Notion backlog tagged by format. The content team stopped starting from a blank page.

None of these are CRM features. All of them are work the team was doing manually, slowly, or not at all. Each one is a small agent shaped exactly around a judgment-shaped gap that no roadmap will ever close.

The buying question changes

If the agent is the interesting layer, the buying question for CRMs changes.

The old question was: "Does this CRM have feature X?" Sales engineers built their careers on that question. RFPs ran on it. Whole industries of comparison content existed to answer it.

The new question is two questions.

First: "What is my team still doing manually inside this CRM that an agent could do?" That question has nothing to do with features. It is a question about hours, friction points, and the work that survived every previous round of automation.

Second: "Will an agent run cleanly on top of this CRM?" That has to do with API surface, MCP support, permission model, schema flexibility, and whether AI-generated fields are treated as first-class. Most of the CRMs in the comparison matrix still fail this second question. A few pass.

You can answer both questions in a week. The answer changes what you buy and what you spend on it. A team that buys a CRM in 2026 without asking these is buying for a category that does not exist anymore.

What this means for the next twelve months

Three things are going to happen, and they are already starting.

The CRM vendors will all ship their own agents. They will be okay. They will demo well. They will not be the best agents for any specific team's work, because they have to be generic to ship, and the value of an agent is in how closely shaped it is to the team's actual judgment.

A small practice industry will form around shipping custom agents on top of the CRM the customer already uses. The economics are good because the build is the expensive part and the runtime is the customer's already. Some of those practices will sell agents that pay for themselves in the first month, because the work the agent does is work the team is already paying for in hours.

The buyers who keep comparing CRMs by feature list will pay roughly the same as before. The buyers who start asking the two new questions will spend the same money in a different place and get more business done. Inside three years the second group will be obviously ahead, and the first group will notice.

The rule

A CRM in 2026 is the schema. The workflows. The auth. The API. It is necessary, table stakes, and finished.

The agent is the layer that does the work the CRM does not do. It is shaped around your team, not the vendor's roadmap. It runs on top of the CRM you already chose. It closes the specific gaps your team is still bleeding hours into.

If you are buying or evaluating a CRM in 2026, the question is no longer "which CRM has the most features." It is "which CRM can an agent live inside, and what work am I going to ship that agent to do?"

That is the only call worth making.

Further reading: I wrote a longer piece on why Attio specifically is the agent runtime and another on how I ship agents without hosting a server. Both are worth reading if this argument lands.