“Layers of abstraction” is one of those phrases that gets thrown around until it stops meaning much. But on a recent episode of Pavilion’s Topline Spotlight, host Sam Jacobs used it to name something that’s actually decision-relevant for GTM teams right now: whether you own your revenue data, or you’ve quietly let a CRM vendor design the architecture for you.
Jacobs put the question to Adam Liska, co-founder and CEO of Airspeed (formerly Glyphic), fresh off a $20M Series A: does Airspeed run its own Snowflake instance, a database the CRM writes to and reads from, or something else? Liska’s answer was candid — Airspeed is still at an earlier stage than that. It doesn’t run its own data warehouse.
What it does have is a system it built in-house that pulls from its main sources of truth: HubSpot for CRM, Airspeed itself for conversation data, and Aluna for revenue management. Because those systems have APIs, Airspeed can extract and analyze the data directly rather than being limited to whatever view the CRM’s own reporting layer allows — sales cycle length, win rates, product usage pulled from PostHog, all assembled into a central interface the team actually looks at.
Liska was clear that this isn’t the end state. Eventually, he expects all of it to land in a single system. But he was equally clear that the underlying principle already matters, even before the infrastructure catches up: knowing how to access and reason over your own data is what makes data-driven decisions possible at all, whether a human is making the call or an AI system is doing some of that analysis on your behalf.
The alternative — leaving Salesforce, or any single vendor, to define the architecture your revenue data lives in — means you’re renting someone else’s decisions about what’s structured, what’s queryable, and what’s thrown away.
Why this comes before anything else
That data question turns out to be the hinge for a lot of the rest of the conversation. Liska described the sequencing Airspeed followed as it moved from founder-led sales to something closer to a repeatable system: data first, then playbooks, then hires — never the other way around. Before building a sales team, he and his co-founder made sure they were capturing everything, transcribing every call and logging every email, both internal and external, from day one.
That gave them a real dataset to reason over before they had reps to hand a playbook to. Only once that foundation existed did it make sense to formalize playbooks — which Liska describes less as a static 60-page PDF and more as a living, hierarchical system that gets updated on a weekly cadence as new conversations come in.
It’s the same logic he applies to engineering. Tribal knowledge that used to live in people’s heads is now documented directly in the codebase, so both engineers and AI agents working on it have the context they need. Liska thinks GTM organizations are headed toward the same thing: encoding what used to be institutional memory into a system that updates itself, rather than a document that goes stale the week it’s written.
What Airspeed is actually building toward
Owning the data is also why Liska resists the idea that AI products should go fully headless — quietly becoming middleware behind a chat interface someone else owns. His argument: a team still needs to be met with the right context at the point of decision — what’s the next best step for this deal, this rep, this manager — and that requires an interface built specifically to carry that context, not just an API another system happens to call.
Asked how he’d categorize what Airspeed is building, Liska calls it “revenue execution” — his term for the layer that sits on top of the CRM and existing revenue intelligence tools and actually acts on what they’ve captured, rather than just summarizing it.
Whether that specific label becomes the industry’s shorthand is beside the point; the underlying claim is that most GTM tooling stops at knowing what should happen next, and very little of it closes the gap to make sure it does.
The bigger bet
Zoom out further and Liska’s thinking here connects to a broader question he’s been chewing on: what organizations look like once AI is doing real back-end work inside them, not just summarizing it.
He points to Mustafa Suleyman’s “modern Turing test” — give an AI agent $100k and see whether it can run a business into profit — as a useful frame for where things are headed, alongside recent commentary from Tom Blomfield on AI-native organizations.
Liska doesn’t expect AI to run companies outright, but he does expect it to take over more of the data structuring, pattern-finding, and knowledge-encoding that currently depends on a handful of people remembering things correctly.
Own the data, build the playbooks on top of it, and let that same discipline extend into how the organization runs — that’s the throughline connecting a Nashville lunch that became a six-figure deal to a $20M Series A four years later.
You can watch the full episode on Pavilion’s YouTube channel, and learn more about Airspeed at goairspeed.com.
