Mike Taylor says because AI lets him produce far more, he now has to protect other people's limited attention. He makes explainer videos and worries about flooding Slack and LinkedIn, so digesting his own output has become a core part of his daily work.
Jack Chang built a tabless browser with an agent pane and peek mode because current models can modify the application itself. He says plugins and extensions never let users reshape software this deeply; now he can build personalized workflows for his own editing and research.
Yannick says AI can bring a product to life with assumptions and make a demo look good. But an architect still has to choose libraries, structure code, deploy it, and make it reliable over time, because demo-ready code often is not maintainable for a growing team.
Yannick's Minga tool turns requirements into a task plan with dependencies, lets agents work in the background, and pings him only for decisions or UI feedback during implementation. He wants it to become multiplayer so pirates and architects can share context on the same project.
Dan argues companies are stuck choosing between chasing every new AI tool and shipping reliable work. A frontier team creates separation: a small group experiments and pushes only proven ideas into products, so the rest of the company can keep executing without monitoring every model release.
Mike Taylor, Every's head of evals, takes qualitative model impressions and turns them into personal benchmarks and explicit tests, such as defining what "sycophantic" means. The goal is to find capability overhang and test models beyond gut feel, then share the benchmarks as an internal product.
Arielle, Every's Frontier Wrangler, organizes experiments so learnings can be synthesized into content, products, and consulting concepts. She also built a dashboard that tracks token spend by person and work category, making the experiment ROI visible so the team can see which experiments are worth continuing.
Dan describes two archetypes: pirates run fast, sloppy experiments; architects compress the mess into maintainable systems before it reaches the broader team. That handoff keeps one-off vibe-coded projects from becoming everyone else's maintenance problem, while still letting proven tools move eventually into the wider company.
Arielle says as models improve, her ability to build systems grows. As an operator, she previously waited for vendors or engineers to add tracking features; now she can create dashboards like the token tracker herself. That shift changes what non-technical operators can own without starting an engineering project.
Katie Parrott is adding audience.md to Compound Writing, alongside voice and style guides. It captures the reader profile and hero's journey, then reviews drafts for audience fit and flags where the piece loses the less-expert reader. This makes the writing tool focus on who the draft is for, not just how it sounds.
Kieran says a frontier team must inspire and listen to the wider organization. Running a hundred experiments without helping some stick just overwhelms people; the team's job is integration, making sure colleagues understand and adopt what works. If experiments never land, the organization cannot benefit from them.
