Let me share something I haven't been seeing many other vendors in the agentic sales space talk about. I think it's worth naming.
If you've sat through a demo of any agentic AI sales tool in 2026 ours included you've probably watched it run an analysis. Usually it's something like a competency gap report. Take the team's recent sales calls, run the AI across them, generate a written analysis with rankings, recommendations, and next steps. The output looks great. It usually IS great. The reps named at the top are intuitively the strongest sellers. The reps named at the bottom are intuitively the ones who need coaching. The narrative tracks reality.
Then look at the small print on the screen. The analysis ran on 15 calls.
Fifteen calls is enough to validate that the analysis is directionally correct. It is nowhere near enough to be operationally useful for a company with a 50-person GTM team that generates several hundred calls per week. The whole organisation collectively logs hundreds of customer interactions in five days. Running an analysis on 15 of them is a vibes check, not a measurement.
So you ask the obvious follow-up. What happens when I run this analysis on 1,000 calls?
This is where most vendors get quiet.
Here's the engineering reality nobody is talking about in the demos.
Large language models have token limits. Even the largest context windows in 2026 max out around 2-4 million tokens for the best models. A single hour-long sales call transcript is roughly 12,000 tokens. A thousand calls is 12 million tokens four to six times the largest available context window.
So you have to chunk. You have to summarise. You have to manage memory. You have to decide which calls get processed against which questions, which findings get carried forward, which findings get aggregated up, and what to do when the model starts forgetting things it processed in the middle of the run. There's a well-known failure mode in LLMs called “lost in the middle.” When a model is processing a large context, it tends to remember the beginning and end well, and forget the middle. At a thousand calls, the middle is most of the calls.
This is a real engineering problem. It is not a marketing claim problem.
I'll tell you where we are at GTM Buddy honestly because I think that honesty is a competitive position in 2026, not a weakness.
Today, our skills hierarchy reliably processes about 100 calls per analysis. That's verified across multiple customer instances. We can stand behind it.
This week, we're validating a few hundred. The engineering is in pilot. The early signals are positive but it's not yet “ship it to every customer” ready.
Our target for the end of Q3 is 1,000 calls in a single analysis, without quality degradation in the middle. That's a real number that the team is working toward, with real engineering work happening on memory management, skill orchestration, and progressive aggregation. It is not a guarantee. It is a target.
That progression 100, to several hundred, to 1,000 is genuinely further than most teams in the agentic sales space have published. I haven't seen any vendor in our category honestly publish their current ceiling. I'd rather tell you ours than overclaim. If you want to see the analysis running on a hundred calls of YOUR data, we can show you. If you want it on a thousand calls today, I will tell you what we can demonstrate and what we can't.
Here's why the volume problem matters beyond bragging rights.
Every CRO I talk to wants two things from AI in their sales motion. They want it to help individual reps in flow that's the easy part, most vendors do it. And they want it to give them a view across the whole motion: where are we losing deals, what content moves what stage, which reps need coaching on what, which playbook actually works in which segment.
That second one is the one that requires scale. You can't see patterns across a sales motion by looking at 15 calls. You see patterns across hundreds. The vendor who can reliably analyse a thousand calls in one run, without losing the middle, is the vendor who can answer the second CRO question.
Right now, that vendor doesn't exist at full enterprise scale. We're getting there. Others are too. The category will mature on the back of solving this volume problem, not on the back of better prompts.
If you're evaluating agentic sales tools right now, here's the question I'd recommend asking every vendor in the bake-off:
“What is the largest call volume your platform has reliably analysed in a single run, end-to-end, without quality degradation in the middle, in production at a real customer?”
Watch for honesty in the answer. The vendor who says “Unlimited” is selling. The vendor who gives you a number with a date is engineering. Buy from the engineer.
Including us, if our number isn't yet what you need it to be.
.png)