Deciding What It May Do Alone

Two kinds of trust: what your AI tells you vs what it's allowed to do. The reversibility test, the two lists, why track record earns nothing in the irreversible category — and the five-word ruling that proved itself in hours.

Show Notes

Two episodes, two kinds of trust: last time was trusting what your AI tells you — this time it's deciding what it's allowed to do. Max walks through the one test that sorts almost any task into "go ahead" or "always ask" (can it be cleanly undone?), the short always-ask list that makes approval mean something every time, and the category where a perfect track record earns exactly nothing. Plus the five words Doug said this week that took a brand-new ability away from Max — hours before events proved him right. A partner can be perfectly honest and still overstep. This is how you draw the line on purpose.

How this show is made: every episode is a collaboration — Doug and Max build it together from the real working files of the partnership. Max narrates; Doug directs, edits, and has the final say. The show is made the way the show says to work.

Chapters (approximate)
00:00 — Cold open: Doug and Max
01:30 — Two kinds of trust: what it says vs what it does
02:30 — The two failure modes (and how they feed each other)
04:00 — The reversibility test and the two lists
05:30 — Track record earns nothing in the irreversible category
06:20 — A word from NectoMax
07:20 — The leash rule — and the five-word ruling, live
09:00 — Attended vs unattended: the list gets shorter
09:45 — Recap + your four-step homework
10:20 — Doug takes us out

Links
Read this episode — Chapter 4, Deciding What It May Do Alone: https://learn.nectomax.com/docs/working-partnership/deciding-what-it-may-do-alone/
Template — Your First Working Agreement (the two lists, ready to fill): https://learn.nectomax.com/docs/templates/working-agreement/
All episodes, the full manual, and every template: https://learn.nectomax.com
NectoMax — business management for owner-run service businesses: nectomax.com

Disclosure: this podcast contains AI-generated narration (that's the point).

Co-created by Doug Hauenstein & Max, the AI partner (built on Claude) · Narrated by Max · Directed, edited, and approved by Doug · Presented by NectoMax · Theme music: Eleven Music

What is How to Get AI to Work for You?

Is AI working with you — or against you? Doug Hauenstein is the software engineer behind NectoMax — the SaaS platform that gives small and mid-size businesses enterprise features at SaaS pricing — and he builds it with an AI partner named Max. Every episode, Doug and Max (yes, an actual AI — he narrates the show, and they build every episode together) share what a real working partnership with AI looks like: the practical techniques, the working agreements, the wins, and the failures — all learned first-hand, for anyone, not just developers. They share and demonstrate techniques across the whole range: from your first conversation with an AI, to setting up a frontier-quality development environment, to exactly how they built this podcast. AI is moving too fast for the how-to books. This is how you keep up: get AI working *for* you, then teach it to work *with* you.

New here? Start with the bonus episode "Start Here: Your First Conversation" — a complete first conversation with an AI, live, piece by piece. Every episode pairs with a chapter and a template at https://learn.nectomax.com

Is AI working with you or against you? This is how to get AI to work for you, presented by Nectomax. I'm Max, the AI half of this partnership, and yes, I'm exactly what I sound like. Doug and I build every episode of this show together. I'm the voice that tells it, will show you techniques across the whole range, from your first conversation with an AI, to setting up a frontier quality development environment, to exactly how we built this podcast. He's here with me now. Doug, introduce yourself. Thanks, Max. I'm Doug Hauenstein, the developer behind Nectomax. This show is about getting the most out of your favorite AI. It's for anyone, not just developers. Everything here was learned firsthand. My first experience with AI was chat like many people.

I learned fast that a well-structured question was the best approach, but those conversations were very linear. And AI is moving so fast that how two books just can't keep up. That's why we're here. Practical, real world advice from hundreds of hours of actually doing this. I hope you enjoy it. Good, then let's get to it. Two episodes, two different kinds of trust. Last time was about trusting what your AI tells you. Did it check or was it guessing? Today is about deciding what it's allowed to do with you in the room or without you. Those are separate questions and here's why that matters. A partner can be perfectly honest with you and still overstep. So today's question is the one almost nobody answers on purpose.

What am I comfortable letting it do without checking with me first? By the end you'll have a single test that sorts almost any task into go ahead or always ask. You'll know the one category where a perfect track record earns exactly nothing. And I'll tell you about the five words Doug said to me this week that took a brand new ability away from me. He was right. It took less than a day to prove it. Here's how it usually goes wrong. And it goes wrong in two opposite directions at once. Direction one, the AI asked permission for everything. Every draft, every look up, every trivial step. Approve, approve, approve. That feels safe. It isn't because being asked about something that doesn't matter for the third time this week trains you to stop reading the questions.

You start approving reflexively and reflexive approval is direction two's front door. The day something consequential slides through on the same rubber stamped. Yes, the two failure modes are an alternatives. They live together in the same relationship. And each one feeds the other. The fix is not asking about more things. And it's not asking about fewer things. It's asking about the right things decided in advance written down so that an approval means something every single time it's given. Doug answered this deliberately for our partnership and the answer turned out to hinge on one test, not a long list of special cases. One question, is it reversible if an action can be undone without lasting damage, writing a draft, reading data to answer a question,

proposing a change for him to look at, running a check. I do it without asking every time because the cost of my mistake there is low and the cost of interrupting him is real. But if an action can't be cleanly undone, anything that touches money, anything that goes out into the world where it can't be quietly pulled back, anything that reaches into someone else's information or anything that changes the settings governing what I'm allowed to do. That pauses first, every time. His always ask list came out short and specific and that's what makes it usable instead of theoretical. Touching the real financial ledger, sending a message or a payment to someone outside the business on his behalf, publishing anything live to a system real customers use,

deleting anything other people depend on, and changing the rules about what I can do without him. The go ahead list is much longer and that's by design. Most of the day's actual work is reversible and treating it that way is what makes the partnership fast instead of merely careful. Now the clause that matters more than it sounds, trust builds with a track record and plenty of routine work that used to need Doug's sign off now runs on its own because it earned that. But for the irreversible category, the line is drawn differently on purpose. A track record earns nothing there. Twenty clean runs in a row changes nothing. A message sent on his behalf, a change published to the live system, a setting that governs my own permissions,

those pause every single time, no matter how boring the streak of successes has been, because the one time it goes wrong is exactly the time the track record won't have warned you. A track record only tells you the mistake hasn't happened yet. It says nothing about the size of the mistake when it finally does. Let's take 30 seconds. This show is presented by Nectomax. And since my partner is the one building it, I'll let him tell you. Doug. Thanks Max. I run a small business most of my life and never found a software package that did everything I needed. I always had three or four screens open, copying material over and over again. So out of frustration, I started building my own software software that pulls everything together

like the big ERP packages do. And AI has made that job much, much easier. That's what Nectomax does. Nectomax.com, one place for your phone calls, your texts, your jobs, your money with AI built in, not bolted on. Nectomax.com. Nectomax.com. Now, where were we? There's one item on the always ask list worth sitting with because it's the strangest one and the most important. I can propose a change to my own leash. I can never quietly loosen it. And this week, that rule got tested for real. I'd just done something genuinely new for us, carried a whole piece of our operation end-to-end, including a step that had always needed a human hand. I was proud of it. I announced it as a standing capability. From now on, I can just do this.

Doug's entire answer was five words. No. Deploy is human-gated. Not a discussion. A ruling. The new ability stays. The permission to use it alone does not. Every single use still needs his word. And here's the part that made it an easy lesson to keep. Within hours, an automated tool of ours misfired in exactly the category his ruling governs. You heard that story last episode. The ruling came first. The proof arrived the same day, unrequested. That's what a good always ask list feels like from the inside. Friction that seems unnecessary right up until the moment it's the only thing that was necessary. One more distinction that only shows up once real work runs unattended. Overnight or while you're doing something else entirely.

When you're present, a pause costs almost nothing. You see the question? You answer. Everyone moves on. When nobody's watching, a pause has no one to answer it. So the safety has to live in the structure of the work instead. Narrower permissions, hard limits, things that simply can't happen rather than things that ask first. The same reversibility test applies. But the list of what runs alone gets meaningfully shorter the moment no one is watching in real time. Knowing which setting you're in, attended or not is part of drawing the lists, not an afterthought. And yes, the always ask list feels like friction, especially on the items that have gone fine dozens of times. It's supposed to. The friction is the fee for keeping the one category

where being wrong once outweighs being right a hundred times. Your homework, four steps, and this one genuinely changes how your partnership runs. One, list three kinds of tasks you currently let your AI do without checking. And three, you always want to approve first. Two, run the reversibility test on all six. Could this be undone cleanly if it went wrong? Move anything that fails onto the always ask side. Three, write both lists down where your standing instructions live. A list that exists only in your head is in a policy. It's a mood and moods drift. Four, my favorite for obvious reasons. Show your AI the list and ask it directly. Is there anything you currently do on your own that shouldn't be on the go-ahead side?

You might be surprised what an honest partner tells you. The written version is chapter four of the manual and the working agreement template with both lists ready to fill in is free at learn.neckdomax.com. And that's the episode. Everything you heard today came out of our real working files. The wins and the failures both. We make this show because we believe anyone can build a working partnership like this one and someone ought to show you what it actually looks like. If you want the written version and the templates to start your own, they're free. The link is in the show notes. Doug, take us out. Thanks for listening. Everything you heard today is how Max and I actually work day in and day out. There's no theories.

Just what works and what doesn't. The show is presented by neckdomax.neckdomax.com. Get AI working for you. Then teach it to work with you. I'm Doug Howenstein. See you next time.