Welcome to A.I. Basecamp

I travel from LA to Denver on a regular basis for family trips, and let me tell you — the altitude sickness is real. Any time I visit, there are roughly 3–4 nights of restless non-sleep, which means 4–5 days of exhausted daytime slog.

My point: acclimation is hard. And guess what? A.I. has an “acclimation” curve, too. It took me about 3 months of dedicated slogging across four computers, three A.I. providers, 7–8 models, and three skills frameworks to finally feel like I made a breakthrough in my A.I. workflow.

In my wake, I came out with one amazing website, one powerful app, a massive skills framework that replicated at least a part of my brain, and a more eyes-open, pragmatic approach to harnessing A.I. as a technically “non-technical” professional.

Of course, let’s not forget the three failed apps tossed to the wayside as I rapidly iterated to find my most productive groove! So many nights banging my head against the wall…

Anyway — I’m hoping my acclimation journey is of benefit to you, so you don’t have to go through nearly as much head-banging frustration as I did — or at least you’ll know you’re in good company.

Below are some of the key takeaways I learned from tackling A.I. head-on — with a few less hairs on my head in retrospect — that may help you reach what I’m calling “A.I. Basecamp” yourself.

A.I. expands your output. It doesn’t replace you.

When I hear the talking heads of news and social media declare that A.I. will replace human labor, my eyes involuntarily perform a Liz Lemon-worthy eye roll. That’s the “scary” headline that gets your attention, but the truth is a different picture entirely.

When I started with A.I., I thought it would replace complex decisioning workflows and lighten my mental load.

Wrong.

A.I. expanded my output capacity — specifically content. I could produce more documentation, draft more Slack messages, scope more tickets, write more campaign copy — but I still had to manage all of it. In fact, A.I. gave me more to manage. My mental load went up as I wasn’t just managing a team of humans anymore — I was also managing a slew of A.I. agents and their prolific, exceptionally verbose output.

My job didn’t get easier, it turned out, it just changed shape.

Here’s how I’d put it: A.I. is the moment we stopped “rowing” and started “sailing”.

Before agentic A.I., we had human-powered teams doing the heavy lifting — creating ad campaigns, building retention flows, strategizing A/B experiments, all the day-to-day work we do as marketers. Now some of that can be delegated. Repeatable processes, basic copy, simple decisioning, rudimentary design — draft work, really — A.I. can handle. But it still needs your constant supervision and QA. The “sails” may be propelling the ship, but humans still need to manage and maintain it.

(Side note: on Lenny’s podcast, Boris Cherny — creator of Claude Code — likened his invention to the printing press. I call bullshit. Love Boris, never met the guy, highly respect his contributions — but this is where we diverge. The printing press replaced scribes. Sailing didn’t replace rowers; it changed what they were doing all day. That’s the shift I see in my own work.)

You’re now delegating, directing, and managing where you used to queue up and complete projects yourself.

A.I. is book smart and professionally dumb

My first internship was in demand generation, and I had the wonderful luck of working for two very patient professionals — a director and a CMO — who recognized my ambition and let me draft marketing and sales copy that would eventually land on landing pages, sales emails, and whitepaper downloads.

I remember being required to run every draft past the CMO, who would absolutely destroy my copy with redlines and changes. I was smart, but inexperienced, and they treated me as such.

That’s A.I. today. It’s the smartest intern you’ve ever hired. It has the knowledge of the world at its fingertips…And it’s an absolute idiot at applying that knowledge professionally. It doesn’t (and can’t) carry the years of context and experience you’ve curated that affects your decision-making. It doesn’t have taste.

Would you trust your promising-but-inexperienced intern with admin-level access to your marketing platform? Of course not! So treat A.I. the same way — as a smart, inexperienced intern. Check and re-check the output. Treat everything as draft content.

This isn’t hypothetical. Have you seen the A.I. slop infesting ads, social media, and LinkedIn lately? That’s what unsupervised A.I. produces. Don’t be the source of it.

A solid plan is worth its weight in gold

In high school chemistry, I had a knack for logically figuring out chemical formulas in my head to deduce the answers to questions. It drove my teacher nuts. “You have to prove your work!” she always said.

She wasn’t wrong. Sometimes we just know how to do something — through innate aptitude or repeated experience. But knowing the process behind how and why we do a task just became a thousand times more important.

Remember how A.I. has all the knowledge and zero experience? That means it doesn’t really know the way to get the output you’ve asked for from the input you’ve given. Worse than an intern — A.I. pushes ahead and makes assumptions and hallucinations when it hits a roadblock, instead of stopping and asking for help.

So A.I. benefits enormously from a well-researched, phased plan. When you give it a task or project, think about how to divide that task into bite-sized chunks. Think about how you decide whether an output is acceptable. What are your decision criteria? Challenge yourself to prove your own reasoning, then communicate that to the A.I. That knowledge and experience inside your brain needs to be documented for the A.I. to work with — yes, even the “common sense” parts.

Here’s what that looks like in practice: “For our welcome series, an acceptable subject line is under 50 characters, uses ‘you’ or ‘your’ in the first three words, and doesn’t open with a question.” That’s the kind of thing the A.I. won’t know unless you tell it. It’s obvious to you. It’s invisible to the model.

If the output is only as good as the prompt, the prompt is only as good as the plan. So maximize your effort in planning ahead of execution, not during it.

(If you want my full 7-step planning loop, it’s at the end of the post. For now, the only step that matters is: write the plan down before you start.)

I like to record project ideas on a voice-to-text app and paste the transcript into A.I. Here’s a prompt I reuse all the time:

Turn this project idea (or transcript) into a planning document. Identify areas that need research, then break the plan into digestible phases. After each phase, stop and let me approve before continuing. Once I approve, update the master plan and give me a prompt I can paste in to kick off the next phase with the same quality standards.

Your app idea is probably a waste of time

As you start working with A.I., you’ll see and hear a lot about people building full-blown apps with A.I. agents. Maybe you’ve got a killer idea you want to put out there yourself.

Look — I don’t want to burst anyone’s bubble, but app development is still hard. Product management is a whole profession that requires a ton of time, effort, and attention. And building applications with login flows and user management features is a security risk, especially when built by bullish vibe coders.

My advice: unless you’re actually a developer — don’t waste your time.

Instead, pare your app idea back into digestible automations, tools, or single-purpose workflows (see the appendix). Save repeatable prompts. Keep each one focused on one job. This will save you so much headache, I promise.

The best way to learn is to get started

A.I. can be daunting. My own journey started in the command terminal — yeah, the scary black box that looks like a computer from the 1980s — and the first thing I ever typed was What can I do here? Three months later, that’s where most of my best work happens. But that’s me. You don’t have to start there.

Here’s what your first 30 minutes can look like:

1. Open Claude or ChatGPT in a browser tab.
2. Pick one task you do every week. Just one — subject lines, campaign briefs, performance recaps, whatever’s already on your list.
3. Paste in real context: the actual brief, the actual data, the actual brand voice notes.
4. Ask for a first pass.
5. Don’t accept it. Tell the A.I. what’s wrong with it. Make it try again.
6. When you finally get something useful, save the prompt. That’s your first workflow.

That’s it. You’re using A.I. for real. No terminal, no skills framework, no apps — just a saved prompt and an honest critique loop.

Cut to three months later — two apps, four machines, seven models, tons of headaches, yada yada yada — and you’ll have your own version of this post to write. You’ll ask stupid questions, make stupid mistakes (not in production!), and learn hard lessons. Everybody does.

Here’s the thing, though: this post is a guide to reach the proverbial “A.I. basecamp,” not the summit. God, we’re nowhere near summiting! The work right now is just getting to a point where we feel comfortable using AI in our day-to-day work streams — the point where A.I. stops feeling foreign and starts feeling like a tool you actually know how to use. That’s the climb in front of you.

You’re not alone. See you up there.

Appendix:

My 7-step A.I. project loop

For when you want the long version of the planning advice:

1. Scope — what is the project, what does it entail, why does it matter?
2. Discover — what are your known unknowns? What needs confirmation? Where are the edge cases?
3. Research — clarify ambiguities, resolve known blockers, outline key decisions, gather required documentation.
4. Plan — break the work into digestible phases. If you’re feeling lucky, add execution waves to each phase.
5. Verify — check the plan for pitfalls and errors. Ask: does this actually accomplish what I set out to do?
6. Execute — launch the A.I. against the plan, one phase at a time. Check each output.
7. QA & approve — review the output against your decision criteria from step 1. If it passes, update the master plan and move on. If it doesn’t, send it back with specific feedback (not just “redo this”).

Two notes:

(1) I use A.I. at every step of this process — but I’m the one guiding it.

(2) I record every plan and store it somewhere I can find it again. That way you can pause work and pick up where you left off.

What “single-purpose workflow” actually means

Say you write subject lines for your email program. Instead of dreaming up a “subject line generator app,” build a saved prompt you can paste into Claude or ChatGPT every Monday:

You are a senior email marketer at a [B2B SaaS / DTC apparel / fintech] company. Our brand voice is [3 adjectives, e.g. “warm, direct, no-jargon”]. Below is a campaign brief.Generate

subject line variants (under 50 characters, no emojis)A matching preheader for each (under 90 characters)One A/B test hypothesis explaining which variant you’d test first and whyFor each variant, label the angle (e.g., curiosity, urgency, benefit, social proof). Flag any subject line that uses words my ESP might catch as spam.[Paste campaign brief here]

That’s it. One prompt, one job, plugged into your existing workflow. No app. No login system. No security surface. You iterate on the prompt itself over time — adjusting based on what actually performs in your ESP — and it gets sharper every campaign.

The same pattern works for: turning customer interview transcripts into quote pulls, turning competitor landing pages into teardown notes, turning a long Slack thread into a written decision memo, turning a brief into a creative testing matrix. One prompt, one job.

Glossary

A.I. agent — An A.I. that doesn’t just answer questions but takes actions on your behalf: writing files, sending emails, running code, browsing the web. “Agent” implies the A.I. is doing work in the background, not just chatting with you.

Agentic A.I. — The category of A.I. tools built to act as agents (see above) rather than respond conversationally. Claude Code, Cursor, and similar tools are agentic.

Skills / skills framework — A reusable bundle of instructions, examples, and reference material that you give an A.I. so it does a specific task your way. Think of it as the difference between hiring a freelancer and hiring a freelancer who’s already read your brand guidelines, voice doc, and past campaigns.

Subagent — A secondary A.I. agent that your main agent can delegate to. Useful for breaking a big task into parallel chunks (e.g., main agent plans, three subagents each handle one phase).