The Frontier Team: How to Separate Innovation from Execution

The core idea behind a Frontier Team is to institutionalize focus. By creating a dedicated, isolated unit for AI experimentation, companies can resolve the tension between maintaining daily output and exploring disruptive new technologies. Innovation is not a general organizational duty. It is a specialized function that requires pirates to explore and architects to stabilize. Organizations that fail to create this separation suffer from a dual failure: they either ignore the frontier until they are disrupted, or they chase every new trend, leading to a state of constant, unproductive distraction. This model provides a competitive advantage by allowing a company to move at the speed of the frontier while keeping the core business stable, predictable, and sane.

The Anatomy of a Frontier Team

The primary challenge for modern companies is that AI capability is evolving faster than operational capacity. The Frontier Team resolves this by creating a separation of concerns. Rather than forcing every employee to keep pace with every model release, which leads to burnout and fractured attention, the Frontier Team acts as a filter.

> I think it lets us move incredibly quickly on what is new without distracting the whole team. And that has a deeply, it is like confidence building for the team because you feel like, hey, I am going to know when things change, we are going to be able to handle it.
>
> -- Dan Shipper

Pirates vs. Architects: Balancing Velocity and Stability

The team operates on a dual-archetype structure. Pirates are tasked with high-velocity, messy experimentation. They test a billion things to find what works, accepting that most will be discarded. This is the go wide phase. However, speed without structure is toxic. That is where the architects enter. They take the chaotic, vibe-coded experiments and compress them into reliable systems.

This dynamic prevents the common trap of demo-ware, which refers to projects that look impressive in a meeting but collapse under the weight of real-world operational requirements. By explicitly defining these roles, the team ensures that innovation is not just a series of experiments, but a pipeline for product improvement.

The Hidden Cost of Context Saturation

A key insight shared by the team is the burden of communication. As individuals become more AI-enabled, they naturally produce more output. If left unchecked, this creates a secondary problem: the internal flooding of colleagues with new ideas, tools, and workflows.

> I am like regularly accused of saturating the company LinkedIn where like every channel I am posting in, it is like Mike, Mike, Mike and I am really conscious of that so creating that video, you know, creating more videos like that making the work that I am doing more digestible, I think is now actually my job.
>
> -- Mike Taylor

The lesson here is that as you increase your personal productivity via AI, your job description shifts. You are no longer just an individual contributor; you are now a curator responsible for protecting the attention of your teammates.

The 18-Month Payoff: From Vibe-Coding to Infrastructure

The team identifies a clear evolution in AI capability: from simple code generation to agentic orchestration. Early experimentation was limited to single prompts. Today, the team is building systems where agents manage dependencies, break down complex plans, and execute tasks autonomously.

The competitive advantage lies in the willingness to build power tools that solve specific, personal bottlenecks, like Jack Cheng’s tabless browser or Katie Parrott’s Audience.md plugin. These tools are often dismissed by traditional software development cycles as too niche or idiosyncratic. However, by building them internally, the team creates a bespoke operational layer that competitors cannot easily replicate, as it is uniquely tuned to the company’s internal values and workflows.

Key Action Items

• Establish a Frontier Unit (Immediate): Identify 1-2 people who are already AI-curious and give them a formal mandate to experiment. This removes the shiny object distraction from the rest of the organization.
• Implement Architect Oversight (Next 30 Days): For every experimental project, assign an architect to evaluate maintenance and integration. If it cannot be turned into a reproducible system, it remains a vibe-check and is not pushed to the wider team.
• Create Internal Benchmarks (Next Quarter): Follow the Head of Evals model by creating personal benchmarks, saving real, completed work as a test suite for new models. This moves AI evaluation from qualitative vibes to quantitative performance metrics.
• Audit Internal Communication (Ongoing): As individual output increases through AI, implement a digestible content requirement. Ensure that new workflows are documented in video or summarized formats to prevent team-wide information overload.
• Standardize Audience Documentation (Next 60 Days): Adopt the practice of creating an Audience.md file for internal processes. Defining the who and why of a workflow before automating it ensures that AI-generated output remains aligned with company values.
• Shift to Agentic Planning (12-18 Months): Move from using AI as a chatbot to using it as an orchestrator, such as the Minga model. Focus on building plans that agents can execute autonomously, requiring human input only at high-level decision gates.