At the first session of RevGenius Demo Day, Airspeed co-founder and CTO Devang Agrawal took the stage with a deceptively simple promise: to build a prospecting agent, live, in under ten minutes. But the demo was really about a bigger idea. Before joining Airspeed, Agrawal trained large language models on the Gemini team at Google DeepMind, and when he and his co-founder started the company, they kept circling back to one question about go-to-market AI: why do today’s agents keep flying blind?
The problem: agents without a brain
Agrawal’s diagnosis was blunt. “The problem is not the agents,” he said. “It’s the fact that they don’t really have a brain behind them.” Most GTM agents act on surface-level context and shallow signals. They don’t understand what it actually means to sell your product, why customers buy, why they churn, or what separates your best sellers from the rest.
Even when teams try to encode that knowledge into an agent, the result is a static snapshot of the playbook as it looks today, with no feedback loops and no learning from what’s happening in the market. And the market, Agrawal argued, is changing faster than ever. “We probably live in the craziest times,” he said. “The competitive landscape is just changing every single day.” A sales process that can’t keep up starts losing deals it should be winning.
The solution: a commercial brain plus an agent harness
Airspeed’s answer is to solve the intelligence layer first. The company is building what it calls the Airspeed Commercial Brain, a self-learning intelligence layer that stays up to date on its own and connects to all of your customer-facing systems, including every call and conversation your team has.
What makes Airspeed distinctive, Agrawal explained, is pairing that brain with its own agent harness, the execution layer that acts on the intelligence. Together they let teams build agents that don’t just think but actually do the work. With that framing set, Agrawal moved to the live demo.
A command center, not a terminal
The tour started at the agent command center, a centralized view of every agent you and your team are running. Agrawal framed it as an antidote to a familiar problem: people burning tokens on various coding agents with no shared visibility into what’s actually running. A central hub lets everyone inspect agents, understand what they do, and reduces key-person risk.
Crucially, building agents in Airspeed doesn’t feel like living in a terminal. The command center ships with customizable templates, or you can start from scratch. It’s friendly enough that revenue leaders build their own agents without a lot of back-and-forth.
Agent one: prospecting on autopilot
The flagship demo was a prospecting agent triggered by a deal state change, in this case, every time a deal moves to closed-won. On trigger, the agent orchestrates the entire GTM tech stack: it analyzes the personas involved in the win (RevOps, CROs, heads of sales), draws on the commercial brain to understand what actually made them buy, and crafts detailed, hyper-personalized sequences for each persona. It then uses Apollo to find similar people at similar companies and can enroll them automatically.
Teams choose the level of autonomy: fully autonomous, where the agent enrolls prospects and starts sending, or semi-autonomous, where it drafts the sequence and waits for your confirmation. Because the outreach is so tailored to look-alike accounts, Agrawal said conversion rates are markedly higher, generating pipeline both for Airspeed and its customers.
Beyond outbound: the CRO cockpit
Airspeed goes well past outbound. Agrawal showed a CRO cockpit built to answer three questions revenue leaders ask constantly. First, which top deals are likely to slip this quarter, judged not from activity data alone but from every customer-facing interaction, flagging the deals that need immediate CRO attention. Second, which top accounts are at high risk of churning, painted from conversational, product-usage, and ticketing data together rather than the partial picture most CS tools provide. In one example, a large account had opened many unresolved tickets and had a critical meeting looming with no supporting data, a clear signal for the CRO to step in.
Third, and Agrawal’s personal favorite, competitive intelligence. The agents track how often competitors surface in calls and, more importantly, the context. In one case, a major competitor was pricing very differently across segments: aggressive on SMB but pricing itself out at the large-enterprise end. That’s exactly the kind of subtlety a CRO needs to adjust pricing and packaging before losing deals, and it applies to nimble smaller competitors too.
For teams short on time, Airspeed also offers a forward-deployed team of engineers and RevOps experts who will build fully customized agents for you.
Forecasting the LLMs can actually be trusted with
Agrawal closed with forecasting, a topic close to his heart. Airspeed assigns each deal a calibrated closed-won probability rather than lumping everything into a generic best case. The catch: LLMs are bad at math. So Airspeed lets the model write code, executed in a safe sandbox, to do the arithmetic correctly, producing a probability-weighted aggregate forecast you can actually trust. For how the close probabilities are built and evaluated, see Airspeed Forecasting and the deep dive Forecasting you can actually trust.
The offer
Agrawal wrapped with a RevGenius promotion: two months of Airspeed free, an offer he extended to anyone listening to the webinar. The through-line of the whole session stayed consistent from the first slide to the last demo. Give your agents a brain, and they stop flying blind.
