Nobody needs to impress the salesman
The salesperson was never there for the persuasion. They were just there to help with limited working memory.
Exploring the world got ruined by machines
Growing up, it always blew my mind how many places in the world my dad had visited. From Seoul to Constanta to Rotterdam, he had explored so many cities in so many places. And he had stories in each of those places too - the food he tried, the people he met, the sights he saw.
My dad was a marine engineer, which meant he did the engineering work on container ships. And back in the 70s, when a container ship reached port, it took a week to unload, and for my dad that meant a week of exploring a new place was one of the perks of the job.
My dad left the ship life when he had kids, and lucky him because that perk vanished soon after. Ships typically leave port within 18 hours of arriving now because of containerization. The perk is gone, the people working the ships shed a tear, but then the world moved on without any debate. Obviously they weren’t going to hang out at the port if it’s not necessary.
Sales is safe, though
As we all reckon with container-level of disruption to all our jobs, a narrative I keep hearing is that the machines will probably take the thinking jobs, but that the humans will keep the convincing jobs. The writing is on the wall for code, accounting, legal letters, ops, analysis, but the art of convincing another person will have to be done by another human. I don’t think the claim is that the AI can draft legal memos but cannot write persuasion, but rather that no human will want to be persuaded by a machine.
This assumes that the salesperson sells persuasion, but there’s more to it than that. Sales does two jobs: 1) telling people you exist, and 2) helping a buyer who can’t possibly collect and parse all the facts make a call anyway. When a VP Finance is buying A/R software and staring at forty vendors and matching them against three hundred internal requirements1, there’s no chance it’s all fitting in working memory. She’s just gotta pick.
So, she swaps the question she can’t answer - is vendor X the optimal choice balancing our needs and costs and friction to implement - with one she can answer - do I like this sales guy. And this means that the “sales jobs will be safe” narrative is built on safety that was never the salesperson’s, it was borrowed from the buyer.
Will she continue to lend him that safety? Will she even have it to give away? I’m not so sure.
If she’s still the one buying, working memory is becoming less of a constraint. She’s not alone in computing the multivariate mess that is buying an expensive product. More information never disrupted the mediators of a sale, because more information still had to fit in a human head. This wave is different. It doesn’t hand her information. It hands her the decision.
But since the premise of this whole argument is that all the jobs except convincing jobs are being done by AI anyway, it’s not clear the decision will be hers! An agent who is buying has a context window that’s at least 1 million tokens, far outstripping the VP Finance’s working memory, and it can read all the marketing materials from all 40 vendors in eleven seconds, including the parts about the founders developing their life’s ambition of building invoicing software starting in the second grade.
So if an agent is buying, the salesperson better be ready to persuade an agent. But look at what it takes to persuade an agent: content, evidence, benchmarks, case studies, ranking. That’s not sales, that’s marketing. It’s hard to imagine a salesperson optimizing their benchmark tables for tokenization by GPT-9.3.
The salesperson was never the persuasion vehicle. The salesperson just helped you work around your limited context window as a human.
Navigating politics
What if the job wasn’t persuasion, but to be a guide for the enterprise champion to navigate their internal politics to buy the product? For a champion in a large organization to buy something, they have to fight other budget claimants, security review, a CFO. And if the purchase goes poorly, she personally eats it.
So the sales guy helps: he provide a deck that the champion presents as their own. He creates the ROI case. He provides the right peer reference. He gives the champion the 11pm pump up call before the big exec review meeting. He helps her turn “I like this sales guy” vibes into a scoring matrix with an answer to the third decimal place. The sales guy is a co-conspirator.
I don’t think the political campaign part of a purchase is going away - I suspect that for a long time to come, a human signature will be required on a six+ figure purchase. But the salesperson will be less needed than ever. They aren’t needed for the deck, or for the ROI case. Or even the therapy. The politics isn’t going to keep the salesperson safe. The champion wanted to win, she never wanted him.
The last human job is to feel good
Imagine two work dinners happening in 2035. Both at a nice New American restaurant in downtown Manhattan, both over a bottle of expensive wine. One is an Account Executive closing a $2.3 million deal to a CFO. The second is a hedge fund manager who has taken a CFO in an industry she’s covering out for dinner.
In the first dinner, the account executive is trying to make her client like her. She’s provided the deck, the ROI spreadsheets, the references, and now it’s down to her and another product that’s seemingly just as good a choice. The more likeable rep wins.
In the second dinner, the fund manager is making the CFO feel like the most interesting guy at the restaurant. Except, he’s being farmed. He’s going to say something he shouldn’t, but go home pleased with himself nevertheless. The fund manager will expense the wine and make the winning trade. All because a boast needs an audience.
The first dinner won’t happen. The CFO never wanted the salesperson, and now he doesn’t need her either.
The second dinner will happen. The CFO always wanted to show off to the hedge fund manager: he really cares what she thinks of him.
Nobody ever wants to be sold to. Buyers dodge AEs. They treat them with business-appropriate-contempt. The fancy dinner with the AE was not the product. It was a toll on the decision made with imperfect information. Once you remove that, there’s no deep want underneath.
The theory behind why sales was going to be the last job left after AI does the rest of them wasn’t completely unfounded - there’s a kernel at the center of the theory around human relationships that survives, even if the salesperson detail doesn’t. While there was nothing in the sales relationship anyone wanted replaced, there is something about the CFO wanting to look impressive to the hedge fund manager that won’t go away. Nobody ever needed to impress the salesperson in that way.
It’s the same thing that drives my dad’s stories of the places he went to, even though the machines have made it so that others can no longer enjoy that perk. He loves talking about them, and, even when it’s the 427th time, I love hearing about them. And it wouldn’t be the same if he told them to Claude, or if Claude made up stories about visiting Constanta for me.
That connection is a core human desire, and regardless of whether the machines end up caring about us, we’ll continue to care about each other.
The question isn’t whether the job involves another person. It’s whether the other person was the product.
Cash application and payment matching (21), collections workflow and dunning (17), invoice generation and delivery (14), dispute and deduction management (11), credit review and limits (9), customer self-service portal (12), ERP integration (16), reporting and cash forecasting (16), revenue recognition and multi-entity and multi-currency and tax (15), API and webhooks and data export (10), security and access control (20), SSO and user provisioning (8), data residency and retention (9), audit trail and SOC 2 and ISO 27001 and documentary evidence of all three (16), business continuity and disaster recovery (10), vendor financial viability and certificates of insurance (12), implementation and data migration (15), training and change management (8), support tiers and escalation paths (11), commercial terms and price protection (9), accessibility, every vendor attaches a VPAT, nobody opens it (9), sustainability and ESG, added 2023, not scored (12), AI and machine learning capabilities, added 2024, not defined (14), mobile, nobody has ever applied a cash receipt from a phone (5), and requirement 287, “solution must not require a Java Runtime Environment” (1).