Passer au contenu
tl;dr: Will AI replace CRM?
No. AI is replacing manual CRM entry, not the CRM itself. The future of CRM software is a record that AI fills in and people still stand behind.
- Gong’s CEO, Amit Bendov, argued on LinkedIn in July 2026 that CRM field updates will fade and conversations will become the record.
- He’s right that fields lose most of what happens on a call. By his count, the average opportunity form has 600 of them.
- He goes too far on forecasting. He puts AI accuracy at 80 to 85%, and forecasts and commission can’t run on that.
- The safest setup lets AI write what the buyer actually said and leaves stage, amount, and close date with the rep.
AI is not going to make CRMs obsolete. But what is going to become obsolete is manual CRM data entry, and the two often get mixed up.
One of the loudest voices on this is the CEO of Gong, Amit Bendov. Amit posted a LinkedIn post on July 2, 2026, where he argued that AI filling in CRM fields is merely a stopgap. He states, “Every CRM field is lossy compression of reality.” The 93 comments, from a mix of sales and RevOps people, were split, and plenty of them didn’t agree.
So, as someone with a former career in sales, I was always fastidious when it came to filling in the CRM. My brain can be a bit of a sieve, and I know that having all those little details can really help when you’re 3 weeks later, trying to remember what was said. But that was an era before AI, so to really dig into what the feelings were, I read every single one of those comments and the full transcript of his podcast interview with Sequoia.
My thoughts? I think Bendov is right about the fields but wrong about the forecasts.
Full disclosure: I write for tl;dv, which connects to CRMs and competes with Gong. Every claim about Gong in this article links to its source.
What Did Gong’s CEO Actually Say?
Amit never actually said, as far as my research went, that the CRM will disappear. So the “CRM is dead” attribution isn’t accurate. CRM will survive as a database, but the work of using it, updating it, reading it, actually doing something with it, that goes to AI. His July 2026 post then builds on that, saying the fields on a CRM will fade away.
The post actually starts with a really interesting point: “Have AI update my CRM” is one of the most requested features they get, and they have built it. Bendov says this is the wrong thing to ask for. No rep wants to be adding data to a CRM, and that query in itself is simply asking for a “faster horse.” The vast majority of detail that gets fed into the CRM is performative and never actually gets used. Taking the words of a prospect and turning them into a dropdown menu selection strips out what happened.
The thing that I did find insightful was about questions. A field selected can only answer something that someone thought to ask in advance. So, you end up with a ghost town of custom fields that nobody uses. His logic is that we will shift away from the field system, and instead the calls and emails themselves become the record that gets queried directly.
What Amit Bendov has said about the CRM, May 2025 to October 2026
CRM will still exist in 20 years, but sales is moving from a CRM-centric world to an AI-centric one.
On the CRM. Sequoia Training Data podcastGong can update every opportunity record, so nobody needs to update fields. Some companies still want the rep to confirm each change, because the AI could be wrong.
On CRM fields. Same interviewGong reaches a half-billion-dollar ARR run rate. CNBC's summary places it above the CRM as the system of action.
On the CRM. CNBC, with Jon ForttAI filling in CRM fields is a stopgap. Calls and emails will become the system of record, and the field update will fade.
On CRM fields. LinkedIn postA small number of fields still makes sense. The problem is the 600 on the average opportunity form.
On CRM fields. His reply in the same threadCRM was built for tracking customers. Revenue AI needs a system of action built for unstructured data, not forms and fields.
On the CRM. LinkedIn, Gong Celebrate recap
As you can see, Bendov’s view of the CRM itself hasn’t changed much over a year and a half, but his view of CRM fields is the one that’s shifted.
Sources
- Sequoia Capital, Training Data podcast, “Gong’s Amit Bendov: From Meeting Recordings to Revenue Intelligence,” May 20, 2025.
- CNBC, “Amit Bendov, Gong CEO: A Fortt Knox Update,” May 13, 2026.
- Amit Bendov, LinkedIn post on AI updating CRM fields, July 2, 2026.
- Amit Bendov, reply to Aaron Hill in the comments of the same post, July 2026.
- Amit Bendov, LinkedIn post on Gong Celebrate ’26, October 2, 2026.
Gong sitting above the CRM is a pretty reasonable product strategy, if a little bit convenient for the one that claims to sell that specific layer. It’s also a story that suits a company Bendov has told Globes is preparing for an IPO once the market reopens. Both can be true, but the question I’d pose is what AI should be allowed to write once it’s looped into the system.
What Does Gong’s CEO Get Right?
Credit where credit is due, I do think that Bendov is right and that the manual CRM entry is no longer fit for purpose, and that a lot of CRM fields flatten the details that help land a deal into a multiple-choice answer.
When it comes to fields, your reps could listen to a potential customer explaining why they are going with your competition. They’ll cover features, pricing, and relationships, but when it comes to the CRM, it’ll be a drop-down box: “Loss reason: Price.” Amit calls this “lossy compression,” and I think it’s a fair term. The CRM records the outcome but misses the detail unless someone has carefully added notes with it, but it won’t necessarily populate a report.
Then there is the issue of bloat. Bendov states that the average opportunity form has around 600 fields, and even claimed he saw a company with 2,400. There is little chance that all 600 boxes are getting filled in correctly every single time, so the data that matters is getting drowned out by noise.
The CRMs I used to use were a bit simpler, but even then a single box that had customer details was not being pulled through to our sales leader’s forecast reports, and a lot of reps would just mark “x” and move on.
He’s also arguing against a feature Gong sells, and says customers love, which adds weight to his argument.
Where Does Gong’s CEO Go Too Far?
In spite of these insights, Bendov probably goes a little overboard when it comes to forecasting. In his Sequoia interview, he suggests that manual forecasting isn’t needed because AI does a better job and that the sales stages are outdated and no longer needed.
The strongest argument against this comes from Bendov himself in the same interview. His statement is that AI can’t replace people wherever there is a need for accountability or a level of decision-making involved. In the interview, he said that Gong’s own answers are roughly 80% to 85% correct, and that these are accepted because they are fast. He also noted that commission is the kind of calculation that cannot be wrong (as a former salesperson, I concur).
But a forecast is that exact kind of number. It’s a formal declaration that goes to the board, and it decides whether reps get paid.
A forecast needs a number a person stands behind, and by Bendov’s own standard, AI isn’t there yet.
He even went on to say what the right setup was. Anything that’s customer-facing: Gong drafts the email and the person manually clicks send. He called this level four on a five-level scale. So a rep in the loop, but the fields that feed the forecast kind of deserve the same rule.
What Did Sales Leaders Say Back?
Most of the thread agreed about bloated fields. The pushback was about dropping structure altogether, and three objections kept coming up.
Forecasts Need Structured Data
Navin Persaud, a VP of Revenue Operations, commented that running a forecast purely off a call recording can’t be done. His point was that the CRM as a place that reps read may become less of a thing; the data layer matters the most because the AI agents need that structured element to read and write from.
Writing It Down Is How Reps Think
I agree with this one myself, and as one of the most liked comments on the entire chain of replies, I’m not the only one. Roi Talpaz, from Backdrop, a sales tool for technical reps, posed that qualification frameworks such as MEDDPICC are in the CRM to test the rep to see whether a deal was actually qualified; handing that task to an AI defeats the point.
Transcripts Miss What Isn’t Said
I’d never forecast based on just “vibes.” The transcript, even if it’s detailed, verbatim, and correct to the point, doesn’t have the full picture. Charles H., an account director, described a call where an AI review of the transcript said the buyer saw value. The video showed the buyer’s delivery, which showed it wasn’t genuine.
Why Does the CRM Still Have to Exist?
The CRM has to continue in some way, shape, or form because it is a shared record for the whole company, and a lot of what is in it isn’t going to come up in transcripts. A sales team may be able to work entirely out of that, but a finance team or customer success cannot. And when a sales leader needs to see the whole pipeline in 30 seconds before a board call, they need a screen, not a search through hundreds of transcripts.
What Conversation Data Can’t Tell You
Calls and emails hold what buyers said. They can’t tell you your weighted pipeline by region, who owns an account, or when a contract renews and at what value. Those answers need structured fields that someone has set up, often from contracts and invoices that no meeting tool will see. Even connecting everything up through an MCP doesn’t change that, because the data still needs to live somewhere structured. And sometimes the conversations that we are looking at have inaccuracies, dates that won’t happen, budgets that have been fabricated; the CRM is meant to be the source of truth.
If most of your company relies on the CRM, keep it at the center and connect AI to it.
Even Gong Connects to a CRM
Yes, even Gong connects to a CRM. According to Gong’s help center, it connects to one CRM per instance and builds its forecasts on that data. It is able to run without it; a feature called Auto Context builds accounts from data it has, such as calls and calendar events, which suits a team that never had a CRM.
For a fuller comparison, see how Gong works.
What Gets Lost When AI Does All the Writing?
Using an AI-only route loses something in the mix. If AI writes straight into your CRM fields and guesses or lacks detail, it runs the risk of hallucination or misunderstanding. If you get rid of the fields and query the conversation instead, just as Bendov proposes, the record stability shifts, and the same question asked twice can deliver two different outcomes.
So, for example, a buyer says to a rep, “We’d love to get this set up before Christmas, but the budget doesn’t free up until January.” There are two dates here, and the AI may decide that, from the urgency around the holiday period, the close date is December. So the deal lands in the Q4 forecast, the sales leader states that to the board, but the deal was never going to close before Q1.
Having an approval step doesn’t fix it all either. A review of automation bias research by Parasuraman and Manzey found that even experts fall into this trap, and no matter how much you train people, it slips through. If a rep manually approves 40 updates a day, there will still be human error that lets it get through.
Extraction vs Inference
Extraction is recording a fact that a buyer stated. A good example would be “procurement begins in November.” This can then be checked against the recording and the transcript.
Inference is a decision that the AI will make from the call. So “this deal should be shifted to Negotiation.” It has nothing concrete to check against; the model has formed an opinion, but it’s delivered as a fact.
Let AI write the facts a buyer stated, and keep its inferences as suggestions a rep decides on.
Which CRM Fields Should AI Touch?
| Field type | Exemples | What AI should do |
|---|---|---|
| Activity log | Call summary, attendees, recording link | Write automatically |
| Facts stated on the call | Next meeting, competitor named, decision maker | Write automatically, linked to the moment it was said |
| Contact details | New stakeholder, job title change | Draft for the rep to approve |
| Deal stage, amount, close date | Stage, forecast category, value | Suggest only |
| Ownership and commission | Account owner, territory, split | Never |
How Does tl;dv Work With Your CRM Instead of Replacing It?
tl;dv records and summarizes your calls, then sends the notes natively into HubSpot, Salesforce, Pipedrive, and Zoho CRM. The CRM stays the record; tl;dv supplies what was said, and you can choose which parts reach which fields.
What tl;dv Writes to HubSpot, Salesforce, and Pipedrive
You map the sections of a meeting template to fields within your CRM. For example, the “Next steps” section can fill a Next steps field; “Decision maker” can fill a Decision maker field.
After each meeting where the template applies, those notes go into the mapped fields automatically. tl;dv locates the record from the attendees. Zoho CRM works through tl;dv’s Workflows Engine instead, which logs each call as a meeting and adds the notes to the matched contact. Field mapping works on one object at a time, so you choose contacts or deals, not both.
You choose what gets mapped. Map what the buyer said, like next steps or the decision maker, and leave stage, amount, and close date with the rep. tl;dv doesn’t train on customer data and hosts data in the EU as standard.
Why tl;dv Fits the CRM You Already Have
The approach this article argues for is how tl;dv already works. It writes what the buyer said into the fields you choose and leaves the judgment calls with the rep. Gong goes further on forecasting and deal boards, if you want your whole sales workflow in one tool and have the budget for it.
You don’t have to rip out your CRM or sign a revenue platform contract to get there. Connect tl;dv to HubSpot, Salesforce, Pipedrive, or Zoho, choose what it writes, and your reps stop typing up calls from memory at the end of the day. The easiest way to judge it is to try tl;dv on your next call.
FAQs on AI and the Future of CRM
Will CRM be replaced by AI?
What did Gong’s CEO say about the CRM?
Amit Bendov never said the CRM is dead. In a July 2026 LinkedIn post, he argued that AI filling in CRM fields is a stopgap, and that calls and emails will become the record. In the same thread, he agreed that a small number of fields still makes sense.
What is the future of CRM?
CRM is moving from the place reps work to the record underneath. AI is taking over data entry and much of the reading, often through tools that sit on top of the CRM. The structured record stays, because forecasts and other teams depend on it.
Do you really need a CRM in 2026?
If you have more than one rep or a forecast to hit, yes. A CRM holds ownership and pipeline data that never comes up on a call. A solo founder with a handful of deals can manage with an AI meeting tool and a spreadsheet, but that stops working once more than one person touches an account.
