CRM · AI Agents · B2B

ClickUp, Monday and Salesforce Are Betting on AI Agents in CRM. Will It Work?

The whole CRM industry is going all-in on AI agents in 2026 (HubSpot included). Here is what the bet means, and why the experts are split.

The moment: two founders, two opposite bets

The clearest way to understand where CRM is heading is to look at two companies that build work-and-CRM platforms and took the AI-agent idea in completely different directions.

It is tempting to read this as "ClickUp fired people and Monday rehired them." That is not what happened, and the real story is more interesting. Monday did not rehire ClickUp's laid-off staff. Both companies went all-in on AI agents; they just disagree on the point of them. ClickUp treats agents as a replacement for headcount. Monday treats agents as new teammates you onboard alongside humans, even giving them a hiring-style marketplace. Same technology, opposite philosophy of work.

And ClickUp's move did not happen in a vacuum. It landed the same stretch as huge cuts elsewhere in tech, part of a wave that has put well over 100,000 tech workers out of roles in 2026, with AI and "efficiency" cited as partial cover in many cases. ClickUp was just unusually blunt about naming agents as the direct substitute.

So what does "an AI agent in a CRM" actually mean?

Now the slightly more precise version. A traditional CRM automation is a dumb rule: "when a deal moves to Won, send this email." It only does the one thing you scripted. An AI agent is different in three ways:

• It is goal-driven, not rule-driven. You give it an objective ("qualify this inbound lead"), not a fixed script. It figures out the steps.
• It chains multiple steps on its own. Read the email, look up the company, score the fit, draft a reply, log the activity, update the stage, schedule a follow-up, all in one run.
• It uses judgment (from a model), not just logic. It can read messy free text, summarize a call, or decide which of five templates fits, things a static rule never could.

A concrete example: one new lead comes in

Say a prospect fills out a form or sends an email. Here is what a CRM agent does, start to finish, without you touching the mouse:

1. Reads the message itself. It understands the email, no template required.
2. Enriches the contact. It looks up the company online, its size, industry, and key people, and fills those fields into the contact record. (In the industry this is called "enrichment.")
3. Scores the fit. It decides whether this lead matches your ideal customer, and flags it hot, warm, or not a fit ("qualification").
4. Drafts a reply. It writes a relevant first response, picking the right angle for that specific prospect.
5. Logs everything and moves the deal. It records the interaction and drags the deal to the correct pipeline stage.
6. Sets your follow-up. It schedules the next reminder so nothing slips.

All of that happens automatically. A human's job shifts from doing the busywork to approving the important moments, like the actual send.

The jobs people hand to CRM agents

In practice, these are the tasks teams delegate most often, roughly from safest to most sensitive:

• Data entry and record updates — the most common and lowest-risk win.
• Enrichment — filling in missing details about contacts and companies.
• Lead scoring and routing — ranking leads and sending each to the right rep.
• Drafting emails and follow-ups — first drafts a human approves.
• Summarizing calls and meetings — turning a transcript into notes and next steps.
• Handing warm leads to a human — the agent preps, the person closes.

In plain terms: in 2026, "AI in your CRM" increasingly means "tell it the outcome you want, and it does the pipeline work autonomously," instead of you clicking through screens to log, update, and chase.

So do you even need a CRM anymore?

Fair question: if you can just tell an AI to handle leads, why pay for a CRM? Because the AI is only the brain. The CRM is the memory, the shared workspace, and the guardrails. An agent without a CRM is a smart worker with no filing cabinet, no colleagues, and no rules.

• One shared source of truth. The CRM is where the data actually lives. The agent needs somewhere to read from and write to; without a central database, every request starts from zero and nothing is remembered.
• A team can work together. This is the big one. Solo, you might get by with AI plus Gmail. The moment you have a team, sales needs to see what marketing did, the manager needs the whole pipeline, and when someone leaves, the knowledge stays in the system instead of walking out the door.
• Permissions, audit trail, and security. A company needs to control who sees what, log who changed what, and stay compliant. A free-roaming AI with no framework is a security nightmare; the CRM supplies the controls.

Is the CRM's built-in AI smarter than a general chatbot? Usually not — many are built on the same underlying models. Its real edge is access: it is already connected to your data, already knows your permissions, and can act safely inside the system. An outside AI would need you to wire all that up and hand it the keys.

What the big players are actually shipping

Every serious CRM now has an agent layer. They differ mostly in scope, price, and who they are built for.

The shift that actually matters for small teams: the "ambient CRM"

The enterprise agent race is loud, but the change that hits a lean B2B operator hardest is quieter. It is the death of manual data entry. A new class of CRM for SMB and startup teams is built to fill itself in. It watches your email, calendar, and calls, then creates and updates contacts, logs every interaction, enriches company profiles, and nudges the pipeline forward, without you typing anything.

This is exactly the demand you see practitioners voicing all over X and LinkedIn right now:

The tools chasing that exact wish:

• Attio syncs email and calendar to build a CRM that fills itself in, enriching contacts and mapping company structures continuously instead of via one-off CSV imports.
• Clarify markets an "ambient" agent (nicknamed Rep) that observes emails, calls, and calendar events and claims to handle roughly 90% of data entry with zero manual input.
• Clay sits on top as the enrichment-and-outreach engine, turning a thin contact into a researched, ready-to-message profile.

The through-line: the CRM stops being a database you feed and becomes a system that observes your real work and keeps itself current. For anyone who has watched a CRM rot because nobody logged anything, that is the actual promise of AI here, bigger than any single "agent" feature.

What this means for you

If you run a small B2B or SaaS operation, a few practical takeaways:

• Data hygiene becomes your competitive edge, not your chore. Agents act on what is in the CRM. A self-filling, well-structured CRM makes every downstream agent smarter. A messy one makes them confidently wrong.
• You do not need Salesforce to get this. The best CRM for SMB teams here is often an ambient, auto-logging tool aimed squarely at small businesses that want to live in Gmail, not learn an enterprise platform.
• Start agents on the boring, safe work. Enrichment, logging, meeting prep, and follow-up drafting are low-risk, high-relief wins. Keep a human on anything that touches pricing, contracts, or a real send.
• Attribution still has to be right. If an agent is qualifying and routing leads, the source data behind those leads has to be clean, or the agent optimizes toward the wrong channels. (This is where a proper first-touch measurement setup pays off.)

The skeptical case: not everyone is buying it

It would be lazy to present all this as inevitable progress. Some of the sharpest pushback comes from people deep in the software world, not from AI skeptics on the sidelines. Israeli entrepreneur Michael Lugassy (@mluggy) laid out a widely shared version of the argument, and it is worth taking seriously. Paraphrasing his points:

• Cutting 20% at the peak of the hype, to widen margins, is a bet against your own people. Framing layoffs as a bold "AI-first" vision can just be a nicer label on old-fashioned cost-cutting.
• Automating a broken process does not fix it, it entrenches it. For years the promise was that we would fix the messy way work happens. Instead, many teams are wrapping the same broken flows in a layer of hasty automations, built by fewer people, leaving the org with double the dependencies and half the understanding.
• The inspiring narratives serve the narrator. "A home where humans and AI agents work together" is a lovely story pointing in a nice direction, but it is still a story told by someone selling the platform. Picking a vendor on the strength of its story, rather than proof it actually works, is how you get burned.
• Volume is not validation. Thousands of workflows and thousands of companies building "agent studios" tell you the category is hot, not that any given setup delivers. The real question is which promises you can actually rely on.
• The honest advice is "less software, not more." Validate that a process genuinely works by hand first. Only then automate a proven flow. No illusions, no pre-picked side, no automating a train nobody needed just because you can.

This view does not say AI agents are useless. It says the industry is racing to automate before it has proven the thing being automated is worth doing, and that a lot of "AI transformation" is really just organizational churn wearing a better outfit. For a small team, it is a healthy corrective: the goal is a process that works, not a dashboard full of agents.

The honest caveats

This is a genuine shift, but the hype is running ahead of the reality in a few ways worth naming:

• "3,000 agents" is a headline, not a headcount. An agent is a running task, not a person. Comparing agent counts to employee counts makes great PR and muddy math.
• Agents inherit your data problems. Point one at a messy pipeline and it will automate the mess faster. Garbage in, garbage at scale.
• Autonomy needs guardrails. An agent that emails prospects or edits deals unsupervised is a brand and revenue risk. The teams doing this well keep a human approval step on anything customer-facing.
• Lock-in is real. The more your workflows depend on one vendor's agents, the harder it is to leave. Worth weighing before you rebuild your whole motion around a single platform.

FAQ