CHAPTERS
- 0:00 – 0:27
Context over code: the new leverage for builders
The conversation opens with a reframing of what matters as AI accelerates software creation: raw code output is commoditizing, while context, judgment, and customer understanding become the differentiators. Founders emphasize rapid change and argue agents increase—not decrease—engineering value by removing toil and raising the quality bar.
- •Code becomes cheaper; deep problem context and customer relationships become scarce
- •The pace of change is accelerating—teams must adapt continuously
- •Agents remove toil, enabling faster iteration and higher-quality software
- •Engineering leverage rises as expectations and ambition expand
- 0:27 – 1:36
Meet the devtool founders: what each company is building
Founders introduce a wide range of developer and infrastructure products—security, email, conversation data, PR review agents, web data APIs, compute management, documentation, local LLM tooling, and subscriptions. The common thread is building developer infrastructure that increasingly serves both humans and AI agents.
- •Identity & secrets management platform (Infisical)
- •Email API positioned as “Stripe for email” (Resend)
- •Conversation data API for meetings/calls (Recall.ai)
- •Agents for PR review/testing at massive scale (Greptile)
- •Web data API, compute platform, AI documentation, local LLM tooling, and in-app subscriptions
- 1:36 – 3:00
Who is the user now: developers, agents, or both?
Founders describe a shifting customer model: developers remain key users, but agents are becoming both the interface and the distribution channel. This drives teams to design for machine-consumable workflows (CLIs, skills, exportable data) while maintaining strong developer UX principles.
- •Developers as users, agents as distribution—and increasingly as “customers”
- •Building agent-friendly interfaces: CLIs, skills, and exportable data
- •Infrastructure products increasingly serve agent-to-agent interactions
- •Agent adoption changes internal orgs (marketing/sales/devops agents)
- 3:00 – 4:22
Building for agents first: redesigning products, docs, and workflows
Founders explain how agent-native design changes product priorities: parallelizing work, revisiting core interfaces, and making everything accessible programmatically. A key theme is optimizing documentation and product surfaces for agent context windows to reduce integration errors.
- •Agent experience as a first-class interface requirement (not an add-on)
- •Parallel development: moving from 1 feature at a time to many in parallel
- •Ruthlessly deleting obsolete features as the world shifts faster
- •Investing early in agentic infrastructure like sandboxes
- •Docs redesigned to fit agent context windows, improving accuracy and reducing mistakes
- 4:22 – 7:16
Biggest early mistakes: DX misunderstandings and product instincts
Founders reflect on missteps in developer tooling: drifting away from developer “spirit,” over-indexing on polish versus time-to-value, and being insufficiently opinionated or confident. They also describe the tension between listening to users and trusting internal product judgment to avoid cruft.
- •Drifting from developer expectations (clarity, speed, transparent pricing) created mistakes
- •Over-focusing on “perfect UI” instead of fast onboarding and aha moments
- •Being too generic vs. owning an opinionated API/approach
- •Not trusting founder instincts enough led to product cruft and complexity
- 7:16 – 9:22
Do founders still write code? How agents changed the workflow
Founders give candid answers ranging from “I only prompt” to “I push a PR daily,” with many describing a shift toward reviewing agent-written code. A recurring idea is that the total code shipped increases, while human-authored code becomes a smaller percentage—freeing time for strategy, customers, and higher-level collaboration.
- •Some founders mostly prompt/review; others still ship code daily
- •Coding agents produce the majority of production code (often ~90%)
- •Managers can code again without becoming release bottlenecks
- •Founders code to stay close to the product and recalibrate intuition
- •Agents shift effort from typing to verification, review, and direction
- 9:22 – 11:29
Most unexpected AI discoveries: distribution, judgment, and security agents
Several surprises emerge: models can become a growth channel by recommending products; agents can critique roadmaps when given broad internal context; and highly customized security/vulnerability agents can outperform generic tooling. There’s also a striking example of open models being delivered to offline regions, changing assumptions about access.
- •Agents/models recommending products can drive customer acquisition
- •Given enough context, agents can critique strategy and prioritize work well
- •Custom internal agents (e.g., vulnerability detection) can be unusually effective
- •Open models reaching offline countries via shipped computers expands the impact surface
- 11:29 – 14:01
What’s underrated right now: strategy, open models, and non-engineers coding
Founders argue that as execution accelerates, strategy becomes more valuable—and agents can be used to pressure-test thinking rather than replace it. They highlight the rapid progress of open models and a broader shift: coding contributions are expanding beyond engineering to teams like finance and support, implying a fragmented “long tail” of specialized agent interfaces.
- •Strategy is undervalued; agents are useful for critique and gap-finding
- •Open models are catching up and can be more efficient to run
- •“Best coding agent” is becoming subjective; workflows will diversify
- •Non-engineers (finance/support) increasingly contribute code via agents
- •Coding agents likely won’t be winner-take-all—expect many specialized tools
- 14:01 – 15:10
Predictions: durability in a world of one-shot software and parallel builds
The group predicts that even if software can be generated quickly, durable infrastructure layers can still emerge due to path dependence and ecosystem positioning. They anticipate a step-change in parallel feature development, while noting that verification/review remains a bottleneck that needs major innovation.
- •Path dependence can make today’s infrastructure surprisingly durable
- •Fear of “everything gets rebuilt” is overstated in the near/mid-term
- •Parallel development will scale dramatically (from ~8 streams to far more)
- •Verification, testing, and review are the biggest unsolved constraints
- 15:10 – 20:29
What’s next: constant change, autonomy, agent-chosen tools, and “idea-guy” programming
Closing predictions emphasize volatility: forecasting far out is harder, and builders must be ready for continuous shifts. The panel outlines a trajectory toward autonomy where agents choose tools, manage procurement, and handle operations (e.g., incident response), while humans focus on taste, intent, and customer context—leading to more reliable, feature-rich software and higher leverage for engineers.
- •Expect ongoing rapid change; long-horizon predictions are less reliable
- •Context and customer relationships remain hard to automate away
- •Movement toward autonomy raises identity/work questions but enables adaptation
- •Tools increasingly selected by agents—agent experience becomes critical
- •Humans provide taste/intent; agents produce scalable, production-grade code
- •Agents may manage sign-up/payment/access and unlock far higher usage
- •Incident response is a near-term frontier; engineers become more valuable as quality rises
