The rep was never the customer
The CRM was built to watch salespeople, not to help them.
That sounds like a grievance, so let me be precise, because I mean it as a description of the architecture. I've spent a little over twenty years in sales. I've carried the quota, run the teams that carry it, and signed the renewal as a founder because migrating felt worse than enduring. From all three seats the same fact held: every required field in the system exists so that someone who wasn't in the room can feel informed about what happened in it. The software's real customer was always the reader of the dashboard. The person typing was the supply chain.
Once you see that, 25 years of sales software stops being confusing. So does the year Salesforce is having.
The number their own research buries
Salesforce surveys thousands of sellers and publishes the results as the State of Sales. The recent editions put the share of a rep's week spent actually selling at 28 percent. That figure isn't a critic's estimate. It comes from the vendor's own report, and it has sat under a third, edition after edition, through every feature cycle since.
Some of the other 72 percent is legitimate: prep, internal syncs, travel. But a large and stubborn slice of it is the work after the work. You finish the call, then you produce a clerical reenactment of the call. You close a deal over WhatsApp at 11pm and the system finds out Tuesday. You run the demo of your life, then spend twenty minutes reducing it to six picklists and a notes field nobody will read.
If you're a founder, that sits on top of product, hiring, and the investor update. If you're a rep, it's a second job that pays nothing.
How did an entire industry decide this was fine? Four waves.
1999 to 2010: the job gets created
Salesforce's founding idea was right for its time. Move customer data off the spreadsheet on one laptop and into the cloud, where the company keeps it when the rep leaves. I don't begrudge any of that.
The limitation was technical. Software of that era could only hold structured rows, and no computer could read a conversation, so a human had to translate the conversation into rows. A whole profession quietly picked up a second trade: witness the deal, then write up the testimony.
Testimony is the word I'd ask you to keep. CRM data was never a recording of what happened. It's what a tired person recalled, compressed, and was willing to commit to writing, knowing their manager reads it. Testimony drifts toward what the room wants to hear. Every deal has a champion because the champion field is required. Every closed-lost reason is "price" because price was first in the picklist. Twenty years of forecast calls taught me exactly how far the drift goes.
2010 to 2020: the job gets enforced
When the typing didn't happen, the industry's answer was compliance machinery, and it's worth noticing that none of it reduced the typing. Required fields before a stage change. Exit criteria. Mobile apps so you could type from the car. Hygiene dashboards, leaderboards, the Friday afternoon pipeline scrub. I've run those scrubs. I've also been the founder with a team of three for whom the 4pm Friday chase was me, chasing myself.
Follow the incentives and the whole decade explains itself. Vendors bill per seat, so every additional typist was revenue. Every record entered deepened the switching cost, so the hours your team sank into data entry were, quite literally, the moat. The one party who wanted the typing gone had no leverage over the roadmap. Buyers bought dashboards. Reps got homework.
2020 to 2023: the copilot era
When large models arrived, the incumbents pointed them at the form. Call summaries, auto-drafted notes, a sidebar that answers questions about your pipeline. You probably paid for the add-on. And after every call you were still filling in boxes.
A form is a lossy, after-the-fact compression of a conversation, written from memory by someone who would rather be selling. Point the smartest model on earth at that artifact and you get fluent summaries of bad testimony. The model was never the problem. The input was corrupted the moment a human had to reconstruct it.
The fair objection is that software already listens to calls. Conversation intelligence has existed for a decade and some of it is genuinely good. But watch where its output lands: a summary pushed back into the same stage field, the same picklists, the same objects designed in 2004. The recorder became one more tributary feeding the form. The form stayed the system of record, so the job stayed.
2024 to today: agents on top of the form
Then the industry found a new word. Agents. Software that doesn't describe the work but does it. The largest vendor in the category bet its entire story on this, and you've watched that bet struggle in public all year: reliability complaints from customer deployments, a community revolt over features sidelined to force adoption, a stock price that says the market has doubts. The industry's own post-mortems keep landing on the same cause, and it isn't the models. It's the data the agents were given to read.
I take no pleasure in any of that, and the struggles aren't evidence that agents are a bad idea. They're failing for the oldest reason in computing. An agent reading a half-filled form is the confident wrong answer again, except now it moves. A summary built on bad testimony wastes twenty minutes of your reading. An agent built on bad testimony emails your customer.
Two eras of AI, one confession. They keep aiming better and better intelligence at the same corrupted input, because the input is the one thing they can't change. The boxes hold twenty years of their customers' data and all of their economics.
The myth of the lazy rep
Analysts have published CRM failure rates since the dot-com era. Gartner's old number was about half, Forrester's was close behind, the definitions were loose, and the studies are dated, so hold any single figure lightly. What has never been published is the opposite: a study of salespeople voluntarily keeping a CRM current. In twenty years I haven't met that team.
Meanwhile the same reps who supposedly resist technology adopted Slack without being asked, were living in Zoom within a week of lockdown, and lately have been expensing AI notetakers faster than IT can approve them. People adopt tools that do work for them and avoid tools that assign work to them. Refusing an unpaid second job is just good judgment.
What it has already cost you
If you're a founder, you've lived some version of these. The board slide that was really whatever you remembered on Thursday night. The deal that went quiet during a product sprint and was found cold on Monday. The first sales hire who inherited a pipeline annotated with "good call, follow up" and nothing else, which is how the deal came back to you, which is why you're still doing sales. And underneath all of it, the renewal you signed because moving felt worse than suffering.
None of that was a people problem. All of it descends from one assumption nobody has questioned since 1999.
The question 2026 is actually asking
The industry has finally agreed on the destination. Analysts call it the shift from a system of record to a system of action: software that does the work instead of filing reports about it. Every vendor deck now says agents, and on the destination I have no argument.
But an agent is only as good as the record it reads, and the record is testimony. Twenty-five years of it. The race this year has been to automate action on top of an input nobody fixed, and 2026 has spent twelve months demonstrating where that leads. You don't get trustworthy action out of untrustworthy memory, no matter how good the model gets.
Here's what actually changed, and it isn't that models got smarter in the abstract. The original limitation behind the form expired. The form existed because software couldn't read a conversation, so a human translated the mess into rows. Machines can read now. One can sit through the call, follow the thread, catch the WhatsApp at 11pm, and notice a deal going quiet before anyone on the team does. Start the record from what actually happened instead of what someone remembered to type, and the agents on top of it stop guessing. The incumbents keep skipping that step, because taking it means demoting the form that holds their data and their pricing.
Companies don't disrupt their own moat. They decorate it.
So here is the truth
Your reps were never lazy. For 25 years they were made to feed a reporting layer built for someone else, and if you're a founder doing your own selling, so were you. The scarcest resource on every team I've run was attention on live deals, and we spent it on clerical work because the software demanded testimony before it offered help.
You don't need more reps, and you don't need an agent bolted onto the old boxes. You need what goes in to change. Everything downstream, the summaries, the forecasts, the agents, inherits either the truth or the typing.
The form had a 25 year run because machines couldn't read. That excuse is gone, and we've been building on its grave. In the coming weeks I'll show you what came of it.



