A CEO on rebuilding his company AI-native: gates torn down, everyone ships, designers hold the last 10%.
Build like it's 2030. Listen like you mean it. ft. Ned Dwyer (Great Question) | Config 2026 · Ned Dwyer
18 min total·Actually worth watching closely: ~3 min·2 must-watch clips
- 0:12 – 2:55Listen
Opening: is design dead, or is this where differentiation starts
Opens on the anxiety about AI killing design and lands the core claim: any company older than 2023, without an AI-native team and AI-native growth rates, is treated as legacy — customers leave, you can't raise more money and no one wants to acquire you.
The long, slow SaaS ramp of incrementally adding features no longer exists; going incremental won't save you.
Pure spoken argument, nothing critical on screen — fine to listen to while you work.▶ Jump to 0:12Speaker · Ned Dwyer - 2:55 – 6:10Listen
The differentiation formula: velocity x insight
Everyone is building from the same set of superpowers — Claude, ChatGPT, Gemini — so the only differentiation is velocity multiplied by customer insight: velocity without insight is just serving up slop, and insight without velocity leaves you super irrelevant.
Neither half works alone — this is the theoretical base for every organizational change in the rest of the talk.
Mostly conceptual, low visual dependency; listen for the two failure cases on either side of the formula.▶ Jump to 2:55Speaker · Ned Dwyer - 6:10 – 7:14Skim
The models are still accelerating: 8x and 65%
Cites the Anthropic team's Claude Code numbers: code output up 8x since 2023, roughly 65% of it deployed to production with no human intervention, a share rising each quarter.
The right reading of 'AI can't do this part of my job' is 'it can't do that yet.'
The key information sits on a data slide (the visual moment lands at 379s); glance at the chart and hold on to the two numbers.▶ Jump to 6:10Speaker · Ned Dwyer - 7:14 – 10:20Listen
Reset the time scale, and everyone becomes a product builder
Re-estimates in hours what used to be measured in years — a three-to-five-day PRD written in a morning, and better for pulling in customer feedback, usage data, even a prototype — while everyone from support to the CEO builds product directly.
Reset the time scale first and role boundaries loosen behind it; speeding up PRDs is the most copyable first step.
Narrative and internal examples throughout; there's nothing you have to watch.▶ Jump to 7:14Speaker · Ned Dwyer - 10:20 – 13:00Listen
Permissive by design: approvals move to the back
The gates up front come down — support ships fixes directly, the CEO ships small features — with a design system and subject-matter-expert sign-off before go-live as the two backstops. Engineers move up from low-level coding to strategy and architecture, responsible for bringing customer insight as context into the systems.
Swap gating up front for backstops after the fact and you can have speed and quality at the same time.
A spoken walkthrough of process and org design — dense, but it doesn't depend on the visuals.▶ Jump to 10:20Speaker · Ned Dwyer - 13:00 – 15:05Watch
Live demo: customer signal baked into every release
Everything qualitative — support tickets, sales calls, research — flows into the workflow via MCP; on stage he builds a reimbursement dashboard prototype in Figma Make, creates a study from the same surface, and feeds live insight back into the next iteration.
Research is no longer the bottleneck in front of a release; it rides along with it.
The only hands-on demo in the talk (visual moments at 783s and 864s) — on audio alone you miss entirely what the workflow looks like.▶ Jump to 13:00Speaker · Ned Dwyer - 15:05 – 17:48Listen
Close: the last 10% belongs to designers
With AI, anyone can reach the first 90%; what separates a general also-product from something truly differentiated is the last 10% — defining the system, finding the gnarliest problems in your organization, identifying and curating and holding space for taste. Design isn't dead; the job got bigger, more important and more strategic.
The designer's new job is holding the last 10% delta — the differentiation the models can't flatten.
A closing spoken argument, heavier on conviction and point of view than on visuals; worth hearing through.▶ Jump to 15:05Speaker · Ned Dwyer