Sequoia's MAD framework from AI Ascent 2026

Sequoia Capital held its fourth annual AI Ascent in San Francisco on April 20, 2026. The firm says more than 150 founders and researchers attended, including Demis Hassabis, Andrej Karpathy, Greg Brockman and Waymo's Dmitri Dolgov. Sequoia published its summary on May 8. The opening keynote gave founders a compact framework for building on top of frontier models.

AI as a revolution in computation

Sequoia partner Pat Grady opened by framing AI as a revolution in computation. Sequoia's summary paraphrases his analogy: not faster horses, but cars, and the cars have arrived. The point is that AI is a new platform, not an upgrade to existing software.

The MAD framework

Grady's advice to founders building on the labs was to "get MAD." Sequoia's summary defines the three parts:

• Moats from the customer back.
• Design for affordance.
• Exploit the diffusion gap between model capabilities and what Fortune 500 companies have deployed.

Moats built from the customer back

The interpretations below are PacTech's reading of each pillar. This one echoes Sequoia's 2023 essay "Generative AI's Act Two," which concluded that moats sit in customers, not data. In practice, a customer-back moat comes from owning a workflow, holding the system of record or earning trust in a regulated process. Model capability changes every few months. A customer's processes and contracts change slowly. Defensibility comes from the slower-moving side.

Affordance makes the product obvious

Affordance is a design term for an object whose use is clear without instruction. For AI products, it means the user knows what to do without writing a prompt from scratch. Default actions, pre-filled tasks and clear outputs reduce the effort a buyer needs to see value. A blank chat box has low affordance.

The diffusion gap in large companies

The diffusion gap is the distance between what models can do and what enterprises use. Sequoia frames this gap as the opportunity. Each model release widens it, because large organizations adopt more slowly than labs ship. A startup that packages new capability into a product an enterprise can buy and deploy captures that gap.

Other themes from the day

Sequoia's summary reports that Sonya Huang declared 2026 the year of agents and described three ingredients that have come together: models, tools and harnesses. Konstantine Buhler argued that the cognitive revolution will follow the arc of the Industrial Revolution, only bigger and faster. Other sessions covered long-horizon agents, robotics, data centres in space and data efficiency. Videos are on Sequoia's YouTube playlist.

Founder takeaways

• Map your moat to the customer. Write down which customer process you own and why it would be costly to replace.
• Remove the blank prompt. Ship default tasks that produce a useful result on first use.
• Target the gap. List capabilities your buyers' competitors have not deployed and lead your pitch with them.
• Ship on the release cycle. Plan product updates around new model releases, since each one widens the gap you sell into.

PacTech's AI for Founders module uses MAD as a review checklist for product plans.

Sources

1. AI Ascent 2026
   Sequoia Capital · May 8, 2026
2. Generative AI's Act Two
   Sequoia Capital (Sonya Huang, Pat Grady) · Sep 20, 2023