How to Focus on the Long Game: Devang Agrawal on Lodestar

Host Cliff Simon sits down with Devang Agrawal, co-founder and CTO of Airspeed, in New York. Devang went from Siri at Apple to the DeepMind team scaling language models before ChatGPT, then left to build Airspeed. Devang explains why Airspeed built a context graph before agents, how an agent harness on top of it answers “why is my conversion rate dropping?” without burning 100 million tokens, and why 2027 will punish teams that skip the long-term work.

In this episode

• (00:39) Devang’s path from Apple Siri to Google DeepMind
• (01:56) The DeepMind pub quiz that led to Airspeed
• (03:03) Why Airspeed built the context graph first
• (03:55) Putting an agent harness on top of the context graph
• (04:41) A live example: tracing a conversion drop from CRM noise to S1 to S2
• (06:56) Why 100 subagents can review every discovery call
• (08:13) From agent experiments to standardized agents across the org
• (10:13) Governance, enablement, and buy-in as the hard part
• (11:09) Short-term gains versus long-term capability
• (12:25) Entering the US market from the UK
• (14:23) Why a CTO spends so much time with sales leaders
• (17:24) Leaving DeepMind to become a founder
• (18:14) Rituals: bedtime, travel, and mornings with two young boys
• (20:11) What Devang takes from India, the UK, the US, and the Dutch

Read the recap

The key ideas from this conversation are written up in Stop running AI one rep at a time.

About the show

Lodestar is a show about the moments that shape go-to-market journeys and the people behind them: the wins, the mistakes, and the lessons that do not fit neatly into a playbook.