The one thing AI still can’t do in a sales call (and why that’s good news for your field team)
by Feroz Shaikh
August 06, 2026 | 02 min read
Share:
Reasoning isn’t the same as discovery
In his paper LLMs Can’t Jump, researcher Tom Zahavy explores a question that’s becoming increasingly relevant as AI evolves: can large language models actually discover something new, or do they only get very good at recognizing what’s already there?
His answer, laid out in the paper, draws a distinction that matters far beyond AI research labs. Zahavy separates two kinds of reasoning. The first is working within a known framework: given the rules, find the answer. Language models are remarkable at this. Feed them enough data and enough scale, and they get sharper at spotting patterns, applying logic, and drawing conclusions that follow from what’s already known.
The leap machines still don’t make
The second kind is different. It’s what happens when there’s no framework yet, when the existing rules don’t explain the data, and someone has to propose a genuinely new idea to make sense of it. Scientists call this abduction. It’s the leap Kepler made when he guessed that orbits might be elliptical rather than circular, not because the data proved it, but because he sensed it might be true before he could fully justify it.
Zahavy’s finding is that scale doesn’t close this gap. A bigger model is better inside the box. It doesn’t get better at building a new box. Humans, meanwhile, keep making that leap even when the evidence is thin, because intuition and lived experience fill in what data can’t.
What this means for field sales
It’s tempting to file this under “interesting AI research” and move on. But the distinction has a direct parallel on the sales floor, and most GTM leaders have the framing backward.
AI’s real job isn’t inventing the pitch
Ask any regional sales head what they want AI to do for their field team, and the answer is usually some version of: help reps say the right thing. Smarter scripts. Better objection-handling suggestions. An algorithm that can dream up the perfect counter to a skeptical retailer before the rep even opens their mouth.
Zahavy’s distinction indicates that the real opportunity is not to ask AI to replace human judgement, but to amplify it. The best sales conversations will always rely on a representative’s ability to read the moment. AI ensures those moments do not stay isolated. It captures what was discussed, what needs to be worked upon, and helps entire teams improve at a speed and scale that manual coaching simply cannot match.
The blind spot inside every sales organization
That reframing exposes a blind spot most sales organizations already have. The best reps make these small conceptual leaps constantly, and then the moment ends, the visit closes, and the insight walks out the door with them.
Nobody captures what was actually said, what objection came up, or what worked. Without that, a manager can’t tell the difference between “the market said no” and “the rep never found the right angle.” And the organization can’t replicate what its best people are quietly doing differently every day.
Capturing human ingenuity at scale
That’s the operational capability abduction demands: not just a smarter pitch generator, but a way to capture the human leap the instant it happens, before it disappears.
This is where Bizom Marshal fits. Marshal doesn’t try to replace the improvisation happening in front of a retailer. It captures it. Every pitch made, every objection raised and how it was handled, every SKU conversation, gets recorded and structured so it’s visible and coachable.
Real-time coaching insights help reinforce the right behaviours, while the Marshal Field Analysis Dashboard turns individual conversations into team and rep-level learning. The leap a top performer made last Tuesday doesn’t just help her sale, it becomes something the whole team can learn from.
The executive takeaway
The executive takeaway isn’t about AI’s limits. It’s about where the real intelligence in your sales organisation already lives, and whether you have any way of seeing it.
So the question worth sitting with isn’t whether AI can think like your best salesperson. It’s whether you’d even notice if he did something brilliant this week.
If you’re curious what that visibility looks like in practice, Bizom Marshal is worth a conversation.
