Joshua Tiernan

Joshua TiernanSep 4, 2026

When an AI agent picks the wrong tool, the model is rarely the problem. Your agent chooses a tool based on its name and description. That's the whole mechanism. So "update sheet" and "add customer email to spreadsheet" are completely different instructions, even when they point at the same module. Make's own example is the template worth stealing: Tool name: add customer email to spreadsheet Description: adds a new customer email address to the customer contacts spreadsheet. Do not add if the email address is invalid or already exists. The name covers what it does. The description covers when to call it, and when not to. Then name those tools again inside the agent's instructions so the two match. Rename a tool later and forget to update the prompt, and the agent goes back to guessing. The same thinking applies to instructions generally. Write them like a briefing document. Make lists six things to include: every step the agent takes, the tools it calls at each step, the knowledge files it references, guardrails for expected and unwanted behavior, examples of inputs and outputs, and examples of exceptions. One more from their docs that's worth doing: when an agent misbehaves, don't rewrite the instructions by hand. Feed the current instructions, the execution steps, and the output from the Reasoning tab into Claude or ChatGPT and ask for a revision. Full guide: https://lnkd.in/e6s7xRNa

Joshua TiernanSep 11, 2026

There are always too many things to automate. These are the 30 to prioritize (pick the ones that fit your business). Includes automations for: Tally (forms) Airtable (database) Make (automation) Stripe (payments) MailerLite (email) Calendly (booking) Slack or email (notifications) Claude/GPT (the thinking steps) https://lnkd.in/dGYDc5FN

Joshua TiernanSep 3, 2026

Most people building an AI agent in Make should be building a normal scenario instead. Here are the 3 types of automations you should build (and what to use them for). 1️⃣ Standard scenario. Predefined logic that always produces the same output for a given input. Syncing data, processing orders, updating customer records. Adding AI here makes it slower and less reliable for no benefit. 2️⃣ Scenario with an AI app. Predefined logic combined with AI generated content, where inputs and outputs follow set parameters. Translating, transcribing, summarizing. 3️⃣ Agent. Flexible reasoning, judgment calls, variable inputs and outputs. Categorizing support tickets, screening candidates, analyzing market data. The simple test: Choose tasks for your agent that you'd trust an intern to handle. Which rules out sensitive data, high stakes financial or strategic decisions, and anything with strict legal requirements. If you wouldn't hand it to a smart new hire on day three, don't hand it to an agent. Link to full explainer below 👇