Judgement Game
I was Early GTM at Two
High-Growth Companies.
Here’s What I Learned…
I was Early GTM at Two High-Growth Companies. Here’s What I Learned…
Swaraj Agarwal, Founding Ops
Devi Jha, Co-Founder
July 27, 2026
4 min read
Stop reading. Start running plays.
Get access to Cardinal and put these motions in market this week.
Sales Is a Judgment Game
It’s mid-2026, and AI agents are being well adopted across certain verticals - software development is leading the charge by a mile. A big reason why is because coding and other industries like customer support are often treated as intelligence work. They require logic, pattern recognition, and the ability to solve defined problems inside a known system, all converging to the same output. There is usually a right answer, or at least a narrow range of good ones that can be taught to a model.
Sales operates a bit differently.
It’s mid-2026, and AI agents are being well adopted across certain verticals - software development is leading the charge by a mile. A big reason why is because coding and other industries like customer support are often treated as intelligence work. They require logic, pattern recognition, and the ability to solve defined problems inside a known system, all converging to the same output. There is usually a right answer, or at least a narrow range of good ones that can be taught to a model.
Sales operates a bit differently.
The Outcome Stays the Same
In sales, the objective is constant. You want attention, trust, a conversation, momentum, a meeting, and ultimately revenue.
But the path to get there is never fixed. It changes from company to company, and often from account to account within the company.
One prospect may respond to ads, then email, then a follow-up bump. Another may need direct outreach. Another may need a warmer introduction. Another may simply need time. Another may not move unless their agent recommends the next step.
The goal is always the same, but the route is not.
That is why sales is not a fixed process with a fixed answer. It is a creative sequence of decisions made inside changing conditions. Capturing that context is critical.
In sales, the objective is constant. You want attention, trust, a conversation, momentum, a meeting, and ultimately revenue.
But the path to get there is never fixed. It changes from company to company, and often from account to account within the company.
One prospect may respond to ads, then email, then a follow-up bump. Another may need direct outreach. Another may need a warmer introduction. Another may simply need time. Another may not move unless their agent recommends the next step.
The goal is always the same, but the route is not.
That is why sales is not a fixed process with a fixed answer. It is a creative sequence of decisions made inside changing conditions. Capturing that context is critical.
Validation is Hard
There is a second reason this is difficult, and it sits underneath everything above.
In coding, you find out almost immediately whether the work was good. The code compiles or it doesn't. The test passes or it fails. That tight loop is a large part of what makes the work learnable - an agent can try something, get a clear signal, and adjust.
Sales has no equivalent loop. A reply might come three weeks later. A deal might close for reasons that have nothing to do with the sequence that started it, or die for reasons no one on the team ever hears. Silence means nothing in particular: the message may have been wrong, the timing may have been wrong, the buyer may have been on vacation, or the account may never have been real. And the sample size is usually too small to separate signal from noise, especially for enterprise or early teams.
So the feedback is delayed, sparse, and confounded. That makes it very hard for an agent to learn what good judgment looks like on its own, because it is rarely told when it exercised any.
Even when the signal does exist, it is usually not stored anywhere. The rep who figured out that a particular account only moves after a warm intro knows it, but that knowledge lives in their head, in a call they never logged, or in a Slack thread no one reads again. It leaves when they do. So the model is not just learning from weak feedback, it is learning from a fraction of what the team actually knows.
That is why the judgment has to come from somewhere else - from context that is captured deliberately rather than inferred from outcomes.
There is a second reason this is difficult, and it sits underneath everything above.
In coding, you find out almost immediately whether the work was good. The code compiles or it doesn't. The test passes or it fails. That tight loop is a large part of what makes the work learnable - an agent can try something, get a clear signal, and adjust.
Sales has no equivalent loop. A reply might come three weeks later. A deal might close for reasons that have nothing to do with the sequence that started it, or die for reasons no one on the team ever hears. Silence means nothing in particular: the message may have been wrong, the timing may have been wrong, the buyer may have been on vacation, or the account may never have been real. And the sample size is usually too small to separate signal from noise, especially for enterprise or early teams.
So the feedback is delayed, sparse, and confounded. That makes it very hard for an agent to learn what good judgment looks like on its own, because it is rarely told when it exercised any.
Even when the signal does exist, it is usually not stored anywhere. The rep who figured out that a particular account only moves after a warm intro knows it, but that knowledge lives in their head, in a call they never logged, or in a Slack thread no one reads again. It leaves when they do. So the model is not just learning from weak feedback, it is learning from a fraction of what the team actually knows.
That is why the judgment has to come from somewhere else - from context that is captured deliberately rather than inferred from outcomes.
Sales Is a Creative Act
That judgment starts with a person, and it usually starts as a guess.
The best motions are rarely obvious in advance. They get discovered by a GTM person in the field who notices something about how a buyer behaves and tries something slightly different, and they only become repeatable when that context gets passed across the team.
One of our early devtool customers realized that engineers often didn’t want to feel sold to directly; they wanted to feel like they had found the product themselves. So the team used a simple sequence: a one-line email to the engineer first asking them to try the product, and if that person did, a follow-up to the VP of Engineering.
Another vertical saas company customer found that running ads while their GTM team was making calls made the company feel everywhere to SMB buyers, creating familiarity and trust that a single channel couldn’t.
None of these worked because they were universally correct. They worked because they reflected good judgment and context by the GTM person: noticing what changed the buyer’s perception, then encoding that insight into a motion that could be repeated.
That judgment starts with a person, and it usually starts as a guess.
The best motions are rarely obvious in advance. They get discovered by a GTM person in the field who notices something about how a buyer behaves and tries something slightly different, and they only become repeatable when that context gets passed across the team.
One of our early devtool customers realized that engineers often didn’t want to feel sold to directly; they wanted to feel like they had found the product themselves. So the team used a simple sequence: a one-line email to the engineer first asking them to try the product, and if that person did, a follow-up to the VP of Engineering.
Another vertical saas company customer found that running ads while their GTM team was making calls made the company feel everywhere to SMB buyers, creating familiarity and trust that a single channel couldn’t.
None of these worked because they were universally correct. They worked because they reflected good judgment and context by the GTM person: noticing what changed the buyer’s perception, then encoding that insight into a motion that could be repeated.
Why Agents Matter
Agents become useful when they can execute that encoded judgment.
They are not valuable because they eliminate the need for thought. They are valuable because they can carry out the parts of the sales motion that have already been learned and structured by capturing context across salespeople.
That sequencing matters. An agent pointed at an undefined motion will guess, and because the feedback is slow and noisy, it will keep guessing for a long time before anyone can tell it was wrong.
The best use of agents is not to impose one rigid process on every company. It is to help each company store and scale the judgment it has already earned.
Agents become useful when they can execute that encoded judgment.
They are not valuable because they eliminate the need for thought. They are valuable because they can carry out the parts of the sales motion that have already been learned and structured by capturing context across salespeople.
That sequencing matters. An agent pointed at an undefined motion will guess, and because the feedback is slow and noisy, it will keep guessing for a long time before anyone can tell it was wrong.
The best use of agents is not to impose one rigid process on every company. It is to help each company store and scale the judgment it has already earned.
Why We Built Cardinal
Coding and customer support are more intelligence-heavy because the work is more defined. There is a right answer, and you find out quickly whether you got it.
Sales is different. The outcome is constant, but the path is always changing, the feedback is delayed and confounded, and most of the context that would explain what actually worked never gets written down. That combination is why generic sales automation has underdelivered for years. It was never really a sending problem. It was a context problem.
We built Cardinal around the part of sales that actually compounds. Capture the judgement of your best sellers, turn it into motions that can be run and revised, and let agents execute them across every account. Not one universal playbook shipped to every company, but a system that holds each company's own context and gets better as more of it accumulates.
Companies that win are the ones that can turn their context into a repeatable system, and it starts when they are able to design how they sell.
Coding and customer support are more intelligence-heavy because the work is more defined. There is a right answer, and you find out quickly whether you got it.
Sales is different. The outcome is constant, but the path is always changing, the feedback is delayed and confounded, and most of the context that would explain what actually worked never gets written down. That combination is why generic sales automation has underdelivered for years. It was never really a sending problem. It was a context problem.
We built Cardinal around the part of sales that actually compounds. Capture the judgement of your best sellers, turn it into motions that can be run and revised, and let agents execute them across every account. Not one universal playbook shipped to every company, but a system that holds each company's own context and gets better as more of it accumulates.
Companies that win are the ones that can turn their context into a repeatable system, and it starts when they are able to design how they sell.
Stop reading. Start running plays.
Get access to Cardinal and put these
motions in market this week.
Your first agent is waiting
The agents are ready when you are
Stop Reading.
Start Running Plays.
Stop Reading.
Start Running Plays.
Stop Reading.
Start Running Plays.
The best GTM teams sell differently than everyone else.
Design how you sell with Cardinal.
Your first agent is waiting
The agents are ready when you are
Your first agent is waiting. The agents are ready when you are.
Your first agent is waiting
The agents are ready when you are
Stop Reading.
Start Running Plays.
The AI Revenue Platform built for
high-growth companies.
The AI Revenue Platform built for
high-growth companies.
© 2026 Cardinal. All rights reserved.
© 2026 Cardinal. All rights reserved.
