We came to SaaStr Annual 2026 braced for noise about agents. We got it. But underneath the noise ran three conversations worth your attention, and one honest question that kept resurfacing once the demos were done and the real talk started.
Here is what we took away.
1. Agents are everywhere. Results, not so much.
If 2025 was the year everyone started talking about agentic AI, 2026 is the year everyone showed up to SaaStr having already tried it, most of them with a mixed report card.
The shift in the room was striking. A year ago the question was should we be looking at agents? This year it was why aren’t ours working the way we expected? The technology moved fast. The implementation reality moved slower.
The gap between time invested and outcomes delivered is real, and it was one of the most honest threads of the conference. Deploying agents is one thing. Getting them to reliably close the execution gap, instead of opening new ones, is another.
“Agentic” has become a word that means everything and nothing. The question is no longer whether you use agents. It is what you ask them to do, and what guardrails you built around them.
2. Data quality is the unsexy blocker nobody admits to
Every conference demo looks clean. Tidy data, current CRM, sharp outputs. Production is a different story, and this year more people said so out loud.
Data quality is still the blocker nobody raises in a pitch and everybody complains about the moment they go live. For GTM teams the problem runs deeper than most admit: first-party data alone is not enough. Your CRM and call data tell you what happened. They do not tell you who else looks like your best customers, what intent signals are firing right now, or what firmographic context would change how you work an account.
For AI to work in a revenue workflow it needs a rich foundational data layer: first-party combined with third-party sources like Apollo, intent data, and firmographics. Without that you are not running intelligent agents. You are running confident ones. Confidence without context is a liability.
Spray and pray is dead. Every inbox is full. The only way through is to get genuinely specific about who you reach out to and with what message, and that starts with getting the data layer right before anything else.
3. The best answer right now? Forward Deployed Engineers
There was one consistent answer to how do you actually make agents work in production: someone who sits at the intersection of technical depth and customer context.
Forward Deployed Engineers, FDEs, came up again and again as the practical bridge between the promise of agentic AI and the reality of running it inside a live GTM motion. Not engineers who throw work over the wall. Not consultants who arrive with a framework. People who hold the technical architecture and the commercial reality at once.
We work with FDEs at Airspeed, and our customers are seeing the value too. When you are trying to close the gap between what an agent can theoretically do and what it reliably does for a specific team with specific workflows, that human judgment is the difference.
What we took into Wednesday
These conversations did not stop at the conference floor. On Wednesday we co-hosted the Gradient Descending roundtable with Samuel Colvin, co-founder and CEO of Pydantic, to go deeper on agent engineering, specifically forward deployed engineering and how you actually build reliable, production-grade agents teams can benefit from right away.
The thread running through all of it: the companies pulling ahead are not the ones with the most agents. They are the ones who have been most honest about what agents need to work. Clean data, clear guardrails, and people who build at the intersection of technical and commercial.
A note on what to actually look for
If you are evaluating GTM solutions right now, a few things worth pressure-testing before you sign:
Do not just ask whether they offer agents. Ask whether they are truly AI-native, or just wrapping old workflows in new language. In production, the distinction is everything.
Dig into the data layer. Output is only as good as the data going in. Ask which first-party and third-party sources they draw from, and how they combine them.
Check what they add on top of your existing tools. Watch for platforms that promise a lot, then leave you exporting data into yet another AI tool to get something useful. That is two steps for something that should take one, or happen automatically.
Intelligent agents paired with human relationship-building is the answer. Agents alone are not.
It was a great few days. The honest conversations, the ones about what is actually hard rather than what is theoretically possible, were the most useful ones we had. We are looking forward to continuing them.
