AI killed CRM data entry.
That was the easy part.
The invisible-CRM wave is hitting Salesforce and HubSpot at the right moment. But it automated only the chore reps hated, and left the two jobs that actually run a CRM in place.
Intelligence is becoming free. When reading and reasoning over text costs almost nothing, the software whose only job was to hold a readable copy of your work loses its reason to exist. The CRM was that software. It was a place to type a worse version of conversations that happened in email and on calls, because machines couldn't read the originals. Now they can.
So the form, the fields and the login look exposed, and a wave of AI-native tools is going straight at Salesforce and HubSpot. Before cataloguing them, it helps to know what a CRM actually does. The answer is the lens that makes the whole field readable.
The CRM was never one job. It encoded three. The typist is the rep doing data entry. The analyzer is the engine that turns logged activity into pipeline, forecast and scores. The boss is the manager looking at all of it: the forecast call, the visibility, the "is this rep actually working." And the order matters. CRM was built for the boss. Siebel and then Salesforce sold managers visibility and forecast control. The rep was always the unpaid typist feeding someone else's dashboard.
The 2026 movement fired the typist. That's the entire pitch. Octolane says it saves sales teams more than 25 hours a week of data entry. Day, Coffee and Attio all sell the same liberation from the keyboard.
But look at what every one of them kept. The analyzer is intact and louder than ever: Day's agents review your pipeline every morning and flag the deals worth your attention, and a Folk customer describes it perfectly, "I'll go to folk to get my insights, look at my boards, see how my pipeline is doing, but I don't have to tell it stuff to do." The boss is intact too. The category still sells manager visibility and "a pipeline you can finally trust," which is a promise made to the person watching the pipeline, not the person in it.
So "no-CRM" is the wrong name. It's "no-typing." The real customer was always the boss, and the AI version serves the boss better, because the data is finally clean and the rep can no longer dodge it by not logging. They didn't remove the surveillance. They removed the rep's last excuse for escaping it. The panopticon got nicer to sit in and harder to leave.
That's the tell. Every tool in the wave builds a cleaner dashboard for the manager. None of them build anything for the rep, who stopped typing but is watched as closely as before.
And there's a reason none of them take the boss's job. The boss exists to read across everyone, and no individual can point an assistant at a colleague's inbox. Permissions stop them, with privacy law sitting behind the permissions. That wall is why the record exists at all: the rep transcribes the one version of events the company is allowed to share. Hold that thought through the next part. It's what makes the field legible.
Three groups are pulling at the category. The incumbents are bolting agents onto the record and quietly repricing away from seats. The new entrants are rebuilding the record so it fills itself. And a third group does something smaller than it first looks: it makes the data entry less painful, one rep at a time.
| Player | AI strategy & pricing | Why it matters |
|---|---|---|
| The incumbents — defending the record | ||
| Salesforce | Agentforce agents and Einstein scoring, sitting on Data Cloud, a paid prerequisite. Repriced off seats to consumption: $2 per conversation, $0.10 per action, or $5 to $550 per user. Three pricing models running at once. | The default nobody gets fired for buying. But its own agents cut service seats by about 10%, so the seat model has started to eat itself. |
| HubSpot | Breeze agents and assistant, grounded on its Smart CRM. Moved to outcome pricing in April 2026: $0.50 per resolved conversation, $1 per qualified lead. Supports MCP, and shows an audit card for every agent action. | The easiest on-ramp for SMB and mid-market. The AI is grounded on the record, which still has to be fed to be worth anything. |
| Pipedrive | An AI sales assistant that suggests tips and prioritises deals. Still pure per-seat, $14 to $79 per user. No consumption repricing. | Cheap and pipeline-first. The most conservative of the three, with AI bolted onto the same record. |
| The new entrants — automating the record | ||
| Attio | AI-native record that auto-ingests email and calendar, with permissions for agents. Seat pricing from about $29 per user. | Best funded of the wave: $52M from GV, ~$116M total, 5,000 customers. The smartest auto-filling record on the market. |
| Day.ai | Agents with a name and a job description; the human approves. Seats plus usage credits. | $20M from Sequoia. Built by HubSpot's former product lead. Its pitch: "Claude answers questions, agents do the work." |
| Octolane | Self-driving, "a system of actions, not records." Runs standalone or as agents on top of Salesforce. Tiered seats. | YC-backed, 1,000+ teams, mostly Salesforce and HubSpot switchers. Migrates a team off Attio in four minutes. |
| Twenty | Open source. Ships a native MCP server, so Claude or ChatGPT operate the pipeline directly. $9 per user, or free self-hosted. | 44,800+ GitHub stars. One customer cut its CRM cost by more than 90%. |
| The typist's helpers — making data entry faster | ||
| Claude / ChatGPT | The assistant reads a rep's own gmail, Slack and calendar through connectors. Roughly $20 to $30 per user. | A rep asks it to read their week and draft the CRM updates. It fills the record faster, for one person. It can't see the team: permissions stop it. |
| Slack + Google | Summarises a person's own threads and meetings in place. Slack can push fields to a CRM over MCP. | Lighter data entry from inside the tools the rep already lives in. Same limit: it reads only what that person can. |
| Notion / Sheets | A flexible table with AI built in. | The rep keeps a lighter tracker with less typing. Still a record someone maintains by hand. |
| Meeting AI | Sits on a rep's calls, transcribes, and drafts notes. | Turns calls into entries so the rep types less. Sees only their own calls, not the team's. |
Read the table through the three jobs and the disruption shrinks. The incumbents keep all three. The new entrants automate the typist and keep the analyzer and the boss. The helpers only speed the typist, and not one of them can answer the boss's question, because not one of them is allowed to read the whole team. Most of the money here is aimed at the single job that was always the cheapest to remove. That's why the field looks more disruptive than it is.
There's a second problem underneath all of this. Every CRM, old and new, assumes the relationship lives in email and calendar. Auto-capture didn't change that assumption. It deepened it. The whole new generation is a better and better reader of the inbox.
But the inbox is losing the conversation. Email isn't gone, but it is no longer where business happens, and the gap is widest among the people who now do the buying. Texting has overtaken email as the way people want to reach a company. A majority of younger workers pick messaging over email to get things done. Gen Z is far more likely than older colleagues to leave work email unread. And these are not minor accounts: younger buyers now drive most B2B purchasing decisions.
So the auto-capture CRM is getting very good at reading the channel its most important buyers are abandoning, while the real conversation moves to WhatsApp, Slack and chat. A perfect record of a shrinking slice of the relationship is still a partial record. The new tools inherited email as the substrate and built on it harder, at the exact moment that substrate started to thin.
Line all three groups up against the jobs and the channel they read, then add two rows for what a buyer can never check today: who the AI may read, and whether you can audit what it says. The incumbents keep every job. The new entrants drop one. The helpers ease the typist and hit a wall at the boss. The last column is none of these. It's the product worth building.
| The incumbents | The new entrants | The typist's helpers | What's missing | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| The job | Salesforce | HubSpot | Pipedrive | Attio | Day.ai | Octolane | Twenty | Claude/ChatGPT+ MCP | Slack + Google | Notion / Sheets | Permission & Proofthe build |
| The typistLogging calls, emails and deals by hand | Mostly manual | Mostly manual | Manual | Automated | Automated | Automated | You wire it | Speeds it | Speeds it | Speeds it | None |
| The analyzerForecast, pipeline and deal scores | Kept | Kept | Kept | Kept, louder | Kept, louder | Kept, louder | Kept | Personal only | Personal only | DIY | On demand |
| The bossManager visibility across the team | Kept | Kept | Kept | Kept, sharper | Kept, sharper | Kept, sharper | Kept | Can't: permissions | Can't | Can't | Dropped |
| The channelWhere it reads the relationship | Anything (yours) | Chat + email | Paste-in | All + WhatsApp | |||||||
| PermissionsWho and what may read and write | Role-based | Role-based | Role-based | Agent-aware | Per agent | Per agent | Role-based | Connector scopes | Admin-set | Doc sharing | The whole product |
| ProofCan you check the AI's answer | Limited | Audit cards | No | No | Approve step | Approve step | No | Shows sources | No | No | Built-in log |
scroll the table sideways →
Kept as-is, or blocked Handled or dropped Partial or personal-only
One column reads every channel, stores almost nothing, and can prove every answer. Nobody has built it yet.
That last column is the missing CRM, and it isn't a record. It's the permission layer the helpers can't be. An authorisation to read every channel the relationship uses, WhatsApp included; an agent that answers across the whole team, not just one rep's own inbox; and a log that lets you check any answer it gives. It does, with governance, the thing no individual's assistant is allowed to do, and it does it without anyone retyping a word. No pipeline to maintain, no dashboard to log into, nothing built to watch the rep. The trade-off is on purpose: it hands the front door to the assistant instead of owning a screen. That's the point.
The invisible-CRM movement is real, and it lands one hit that sticks: the typist is gone from the new tools, and Salesforce and HubSpot have no answer that doesn't undercut their own seat pricing. But it stops at the one job that was always easiest to take. The analyzer stays, the boss stays, and the whole thing still reads email while the buyers move to chat. The helpers make the typing faster and then hit the same permission wall the record was built to get around. The last column is the one that breaks the wall: permissions to read everything, a record of almost nothing, and proof for every answer. That's the one still waiting to be built.