Last week we covered steipete's six principles for working with AI, and the whole thesis fit on a bumper sticker: just talk to it.
But, once you try this approach, you hit a question fast… what do you say first?
The answer is almost always "plan with me first," and that step is what we’re covering in today’s edition.
Modern AI already has planning built in. You don't have to write plans yourself; you just have to know how to steer the conversation toward a good one.
Below you’ll find the six things to look for in any plan your AI builds with you, plus a single prompt that turns on planning in any tool you use 👇
10 AI Stocks to Lead the Next Decade
AI isn’t a tech trend – it’s a full-blown, multi-trillion dollar race, and 10 companies are already pulling ahead.
These are the innovators driving real revenue, attracting institutional attention, and positioning for massive growth.
Get all 10 tickers in The 10 Best AI Stocks to Own in 2026, free today.
First thing’s first… turn on “plan mode”
Most modern AI tools already have planning built in.
Claude Code and Codex (the coding agents from Anthropic and OpenAI) have a dedicated plan mode. The AI can read and explore your project but can't change anything until you say go.
Claude.ai and ChatGPT quietly think things through before they answer anything non-trivial. That's the "thinking…" you see before a response on harder questions.
Even when there's no plan-mode button, you can just say so:
"Before you do anything, plan this out with me. Don't start the work until I tell you to."
That's it. Modern models hold their fire when you ask them to. The hard part is knowing whether the plan it hands back is any good.
And that’s what we’ll explore next…
1. The Why
What becomes possible after this is done?
A good plan opens with one or two sentences from your perspective as the user. Not "add an export feature" but "after this, I can click a button and download my customer list as a spreadsheet."
If your plan doesn't have a clear Why, ask:
"In one sentence, what becomes possible after this is done?"
The Why is what keeps the AI anchored when it hits a tricky decision halfway through.
With the Why defined, the AI is thoughtfully building toward something.
2. The Open Questions
A good plan doesn't start with steps. It starts with the AI interrogating you about what you actually want: assumptions it's about to make, edge cases worth flagging, decisions you didn't realize you needed to make.
If your AI launches into a plan without asking anything, push back:
"Before you draft the plan, what do you need to know from me? What did you assume? What's still ambiguous? Ask me one at a time."
Then answer the questions in plain English. Out loud if you can (voice input is your friend here).
The AI updates as you go, and the loop tightens.
Catching ambiguity before any work happens costs nothing (and saves a lot of time).
3. The Steps
The sequence of work, in order.
A weak plan says "set up the thing, build the feature, test it." A strong plan names what each step touches and what changes.
If the steps feel vague, push back:
"Make each step more specific. Name what it touches and what changes."
That single push often turns a 30% plan into an 80% plan, with zero extra work from you.
4. The Proof
How will you know it worked?
Most plans skip this entirely, and it's the single biggest reason work gets "completed" without actually doing what it was supposed to.
Coders call this approach test-driven development (TDD), but the idea translates to almost any kind of work.
Quick version: instead of building the thing and then checking if it works, you write down what "working" looks like first as a specific check anyone (or any AI) can run.
Then the AI builds toward making that check pass.
Why this changes everything when working with AI:
The AI gets an unambiguous target. The check passes or it doesn't.
The AI can grade its own work. It runs the check, sees the result, keeps iterating until it passes. You don't have to babysit.
It catches "fake done." AI agents have a habit of declaring victory the moment they stop typing. A failing check makes that impossible.
It compounds across long runs. This is why Codex can run for seven hours straight. Every step has a check. The agent always knows exactly where it stands.
Anthropic calls this the single strongest pattern for working with AI coding agents (their best practices guide is worth bookmarking).
For non-coding work, the same principle holds. Before any work starts, ask:
"What's the specific check that will tell us this is done? Write it first, then build to pass it."
For a blog post, the check might be "hits the three key points, includes a call to action, stays under 800 words."
For a cleaned-up customer list, "no duplicate contacts, all phone numbers in the same format, a count of merged records at the end."
For a sales email sequence, "five emails, each under 150 words, each with one clear ask."
Define the check first, then let the AI build to it.
5. The Decision Log
A running record of choices made along the way, and why.
You don't write this (the AI does) but you have to ask:
"As we iterate, keep a running log of the choices we make and why."
Three reasons this matters:
It explains the work to future-you. Three weeks later, you won't remember why you picked that approach. The log will.
It lets you resume. If you (or another AI session) need to pick up where you left off, the decision log is the on-ramp.
It catches mistakes. When you see all the decisions in one place, the ones that don't make sense jump out.
6. The Progress Tracker
A checklist of what's done, what's in progress, and what's next. Sounds boring, but it's what makes plans actually work at scale.
When OpenAI ran Codex for seven hours straight, the progress tracker is what let it keep going.
Every time it finished a step, it checked the box, looked at the next one, and kept moving. Every time it paused, the tracker showed exactly where to resume.
If your plan doesn't have one, ask:
"Leave a checklist at the bottom with every step. Update it as we go."
With a progress tracker, the plan becomes a tool that compounds across sessions.
Now, you don't have to memorize all six.
Here's the prompt I use to set them up in a single message. Paste it into any AI tool, drop your task into the brackets, and let the AI take it from there. It'll start by asking clarifying questions before any work happens.
I have a task I want help with.
Before any work starts, let's build a real plan together.
The task:
[Describe what you want to do. Be as messy as you want.]
First, ask me clarifying questions one at a time until you have enough context to put together a strong, cohesive plan. Don't guess at what I want. Ask.
Once we're aligned, draft the plan using a test-driven approach:
For each meaningful piece of work, define what "done" looks like as a specific, checkable result FIRST, then build to pass that check.
Save the plan as a markdown file with a checklist at the bottom covering every task. Update the checklist as work gets completed so we always know exactly where we are.
Start with your first clarifying question.
Your roundup of the latest model releases and updates from the biggest AI labs.
Sam Altman walks back his white-collar job loss prediction. Speaking in Sydney, Sam Altman told an audience that the rapid rollout of AI won't produce the widespread white-collar job losses he'd previously predicted — admitting he'd been wrong about the near-term economic impact. Take it with whatever salt you prefer, but a public reversal from the OpenAI CEO on the doomiest part of the AI-and-work conversation is a meaningful tonal shift in how the industry is publicly framing things. (The Information via Bloomberg)
Pope Leo XIV's first encyclical is about AI (and Anthropic shared the stage) The Vatican released Magnifica Humanitas, Pope Leo's first encyclical (the highest form of papal teaching, this one running 42,000 words). It's a call to slow AI development and keep it under human oversight. The notable detail for builders: Anthropic co-founder Chris Olah co-presented alongside the Pope. That's a deliberate signal about which AI lab the Church considers a credible safety partner — and it puts AI ethics in the same category as the church's historic statements on labor and human dignity. (Anthropic)
Anthropic closes its $30B round at a $900B+ valuation. Anthropic is closing its largest round yet: over $30 billion in fresh capital at a valuation north of $900 billion. At that mark, it becomes the most valuable AI company in the world, ahead of OpenAI's $852B from March. Sequoia, Dragoneer, Greenoaks, and Altimeter co-led, with Microsoft, NVIDIA, Founders Fund, and General Catalyst joining. Investors are pricing this on Claude's enterprise traction (KPMG just put it in front of 276,000 employees) and Anthropic's projected first quarterly operating profit. Reportedly the last private round before an IPO targeting October. (Bloomberg)
Advertise with Build with AI
Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.
Interested in advertising? Respond to this email for rates and details.
Until next time,
William Ryan
Editor-in-Chief @ Build with AI
PS: Follow me on X for daily updates and AI workflows.