AgentGTM / Writing
    GUIDES & FRAMEWORKS

    Writing

    Practical, evergreen guides on building with AI agents and GTM harnesses — pulled from the field notes we've found worth keeping around.

    What is a GTM harness, and why use one instead of ChatGPT?

    A GTM harness wraps an LLM in guided workflows and structured outputs built for one job — here's how that differs from prompting ChatGPT or Claude directly, and when it's worth the switch.

    The AI enablement role companies are quietly hiring for

    Samsara, Sonar, GitLab, and others are hiring a role that doesn't have a settled name yet — someone to build a shared AI context layer, wire up existing tools, and keep the stack current. Here's what the job actually involves.

    Skill, agent, or app — which form should you build your expertise into?

    Bottled expertise now takes one of three forms — a skill, an agent, or a vibe-coded app. Here's what each is for, who it's for, and how to pick.

    Custom GPTs vs. AI agents — what's the difference, and when do you use each?

    A Custom GPT is on-demand conversational help you steer with context and instructions. An AI agent is an automated, multi-step workflow that runs without you. Here's when each one fits.

    Agent orchestration, explained simply — build capabilities like building blocks

    "Agent orchestration" sounds complicated. In practice it means codifying one solved capability into an agent, then building the next capability on top of it — here's what that looks like in a real product.

    Stop generating AI images — ask for editable HTML instead

    Asking AI to regenerate an image for one small edit breaks the whole picture, because it's not editing — it's redrawing from scratch. Ask for a self-contained, editable HTML file instead. Includes a copy-paste prompt.

    Why build in Cursor even if you're not writing code

    The advantage of a coding agent like Cursor over ChatGPT isn't the model — it's that you can see and control exactly what context the AI is working with. Here's why that matters even for non-technical builders.

    What I learned building 10+ AI agents used 14,000+ times

    Lessons from building and shipping more than 10 AI agents, including one used 14,000+ times: agent building is product management more than prompting, and specialists — not developers — are often the best people to build them.

    How to use AI deep research to quantify compliance risk in your messaging

    A four-step framework for using ChatGPT's Deep Research to map a regulated buyer's compliance risk, quantify it in dollars, and turn it into specific, credible messaging — with copy-paste prompts.

    Plan before you build — the loop framework for building with AI

    Building well with AI is mostly planning, not prompting. The "loop" framework — objective, metric, boundary — turns one-off prompts into a system that runs itself until it checks back in with you.

    How to turn your expertise into an AI audit agent — a case study

    A one-person design studio cut homepage audits from an hour to 13 minutes by training an AI agent on her own framework. Here's how the agent was built, and how the case study itself was produced as an editable HTML file.

    If your users live in Claude or ChatGPT, an MCP server matters more than your website

    If people spend their day inside Claude or ChatGPT instead of your website, the way to reach them is an MCP server — not a better landing page. Here's what that looked like for one founder who built exactly that.

    How to decide what your AI agent should remember about a user

    Most AI memory stores everything and resurfaces the wrong thing. A better rule: remember only how a user differs from the average user, and only facts that would change a future recommendation. Here's how we built that into Pangea's traveler profile.

    How to write a voice guide so your AI doesn't sound like marketing

    The voice rules we put in Pangea's system prompt: require an opinion, one vivid detail per section, ban hype adjectives, ration filler words like "worth," and personalize without announcing it. Examples included.

    How to test an AI feature across different kinds of users (and why you rewrite rules instead of patching them)

    Build a small internal tester that runs the same scenario through different user profiles, read the outputs side by side, and when the same miss repeats, rewrite the rule that caused it instead of adding another "don't do this" line.

    How to decide where AI belongs in your product, and what to build first

    Ask users where their current AI falls short, map the whole customer journey including the parts you don't own, pick the moments where you can add the most, and put the AI where users already are. Here's how that played out at Pangea.