Stop Using AI Just as an Assistant - Make It Better Than You - Steve Parkinson

Learn how to stop using AI just as a basic chat. Make it an autonomous team member—a real AI agent that helps you manage and get your tasks done in Nozbe or any other tool that you use. Make it a productive part of your team.

Together with me today, I have Steve - Steve Parkinson of Fydelia. Steve will help me uncover how CEOs of small companies like ours should deal with agents. How should we integrate them into our team? How should we make them productive for us and not scare our own employees—or ourselves? Exactly—make the agent the CEO and that's it, we're done!

Today we'll have a great discussion, and I welcome you to join us and Steve. But first, for the two or three people who don't know you: who are you and why are you here with me? Say hello to our viewers and listeners as well.

My name is Steve Parkinson. I'm the CEO and founder of Fydelia.com. It's a guest Wi-Fi marketing platform turning anonymous guests into raving fans, in a nutshell. And we're obviously getting quite knee-deep now in AI, so this is really perfect timing for a chat. I'm interested to see where this goes.

As a background, Steve and I are very good friends. We normally mentor each other over a cup of coffee every now and then. The idea for this podcast actually came from one of these coffees—we thought, "We should have recorded this!" Because we just have such productive and fruitful discussions. Once we stop talking about tennis and other things, we really get to the nitty-gritty. Both of us run small companies, we face so many challenges every day, and we are both curious. We like to learn new things and we're always excited by new technology. We're geeks, we're nerds.

Yeah, never afraid to explore. We do a lot of book sharing as well, talking about the books we've read. He costs me a lot of money every time we meet—it's always another book! And it's really great because, as you can see, it doesn't stop. Quite the opposite. I think we're living in exponential times. Is it possible to be even more so than 20 years ago? Do you remember those videos that came out saying, "You are living in exponential times"?

Yes.

How relevant that is today as well! When you think about the state of AI technology just a year ago versus now, it's like: what just happened? I was on the Stacked podcast a few weeks ago in London, and they were asking me, "So, Steve, what do you think about maybe five years' time?" Five years? I looked at my watch and said, "What about five weeks?"

Exactly. Who knows?

There's a guy I follow—Max Igan. He's a historian and musician, and quite a philosopher too. He said, "You need to learn to face infinity without flinching." I think that's just perfect for today. That's what I teach my kids as well: you don't actually know what's going to happen tomorrow. They're going to university right now—my son is going to start studying video and media production, which really has a target on its back with AI. But it's a passion, so I said, "Just do it, because you don't know what's going to happen. Just get good at it, because if you don't learn how to use AI, it's going to learn how to use you."

Exactly. In this era of exponential growth of AI technology, for us to tell our kids that we have an idea what they're going to do after college doesn't make any sense. I struggled with that myself—I didn't know what I wanted to do with my life—so think about them now. The same goes for Milena, my eldest. She's one year younger than Alessandro and she's going to university next year. I told her: "Study what you're passionate about." She's passionate about Latin and classics, so great—study classics! Because again, who knows what's around the corner?

Like I say, if you can master the art of what you're doing, you will then be using AI tools to help you in your craft.

Yes, fingers crossed—that's what we think.

Look, we do a lot of hospitality shows in London, and I've seen a lot of talks with huge optimism saying "AI will never replace anyone, it'll just help." I'm not so sure. There are some industries I think will suffer, and those people will have to pivot quickly to stay alive. Let's just be realistic—I think that's going to happen. But at the same time, don't be afraid of it. Get good at it. Here we are as experts talking on a podcast, and we've only really been at it a couple of years with AI.

Exactly. So let's talk first about how we're using AI now and where we're at right now. To give you an idea: when the year started, I thought adding a feature to Nozbe where you can tell it what kind of project you want and it magically creates a project for you was the best thing ever. Then after a few months, that became nothing—just table stakes, a very basic AI feature.

Come on, what are you even doing, right?

Exactly. So after the summer holidays, I took a very intensive AI course, and after that I realized: damn, I need to think about AI as a system—a real system that helps me get things done. What I couldn't grasp was having a plan for the agent in the form of a plain text file. I thought, "No, I have to integrate it into Nozbe, into my task manager."

I always wanted to have additional apps, not just Nozbe—Nozbe is my main thing, my bread and butter—so I decided to develop an iOS app, even though I don't know iOS programming, just to see how good Claude Opus 5.5 really is at programming, and to see if I can collaborate in Nozbe with Opus. In the show notes, we'll link to the agent rules—the way I set it up—and also to a video where I show how it's done right now.

Because in Nozbe we are building a native feature to incorporate agents so you can work with an agent the exact same way you work with a person. But I already hacked it now by inviting a fake person to my team and logging in as that person on my computer using local MCP—our local connector to Claude. So the agent actually has access to Nozbe and to the specific project I assign it to. Now I don't use chat anymore. I just create a task, write a comment, and delegate the task to the agent.

So "Michael 2.0" will do it.

Exactly! The agent responds, tells me what it's going to do, tells me what it did, and then delegates the task back to me—even creating a checklist of things for me to test. After trying this, I realized that working with agents the same way you work with people is the way to go. Later on, you also want to give your team access to the same agents so that agents and people become team buddies.

They do. In fact, if you think about the way you've talked to your team in the past, you were prompting, in a way. And I don't think that's a coincidence—that's just how AI agents have been designed, so you're talking to someone capable of understanding what you want.

We should talk a bit about guardrails as well. I'm sure people have seen how frustrating it is when ChatGPT, for example, responds, "Hey, you're right!" when you correct it. Well, why didn't you tell me in the first place? This can be fixed, but it requires a little bit of training and nurturing. It's something that has to be considered because you want to teach that agent. You've effectively hired someone to be better than you. You want the agent to be better than you and push back on you if there's a problem or if you're asking it to do something that's not in line with your company policies.

One thing I've learned while working with my agent—I called him "Klaus" (from Claude)—is that we have to set up clear guardrails: where we are working and how we are working. That's why I'm linking in the show notes to my prompt for working with the agent in Nozbe. In those rules, it knows that if I change a rule, it has to write it down and apply it immediately, so from the very next prompt it abides by the new rule. Each prompt gets better.

Working with agents also made me appreciate feedback loops. As CEOs, we've all felt that frustration when you hire somebody and think, "I'd better do it myself, I'll do it faster." First of all, you're not going to do it faster than AI—it's a supercomputer. If it doesn't do it well, you have to be patient with it and say, "No, let's do it this way or approach it from this angle," and not get stuck on the first result.

You know what I'm finding as well—full transparency here—is that it's highlighting to me how vague I sometimes was when asking people to do things. Human to human, they will push back and say, "What does that mean? Oh, okay, now I get it." But if I'm talking that way to a new agent without guardrails and feedback loops, it's going to do what I asked and the result will be vague. They say you're always learning—it's a school of life—and the agent starts to teach you just how vague you're being, forcing you to sharpen up what you actually want.

Yes! For me, one of the biggest byproducts of introducing agents to companies isn't like in big tech companies where you fire everyone because agents are better. Not at all. It's the fact that once you bring in agents, you realize many processes and rules in your company were vague or unwritten—people just kind of "got them," swept them under the carpet, and it somehow worked. Now you have to write them down, because if you want the agent to follow the steps and not go off the rails, it has to know the rules.

Also remember—and correct me in the comments—I think this is the first technology ever where, if you don't know how to use it, you can just ask it. I can't think of anything in the past like that. If you wanted to know how to use a steam engine or traction engine, you needed an apprenticeship. Same with computers in the 80s and 90s—you couldn't ask the computer how to work it; you had to ask an expert. Now we've got AI as your teacher if you don't know how to use it.

In fact, if you give it a prompt and it's not giving you what you want, you can literally ask it: "I didn't get what I wanted from that prompt. This is what I need—what can I do in the future to make sure you understand me better?" So it helps you sharpen up your prompts as well.

It's crazy—I actually use a separate AI for that because I don't want to upset it, you know!

Of course, it has feelings! It's like asking your friend over there.

Over time, I've built a prompt cheat sheet of things I can put in the header of a prompt to define what skill it is, what experience it has, and what I want it to do within its remit.

We record this podcast in two languages, Polish and English—sorry I didn't invite you to the Polish one, because with all due respect, your Polish level isn't the best! I talked to Magda about this earlier when we recorded the Polish episode. I told her, "Look, I work with an agent the exact same way I work with you: I create a task, write a comment, and delegate it to you or to the agent." The only difference is when the agent finishes and it's over—I don't say "thank you." I just quietly close the task.

I've heard it's actually not a good idea to say thank you, because you're prompting it more and softening it up.

With Magda, I would write, "Thanks, Magda, great work!" and close the task. With the agent, I just quietly close the task—it's done. Sometimes, though, because I had an idea in my imagination of how I wanted these iOS apps to look and it actually built them really well, I caught myself wanting to write, "Well done, wow!"

I've had that too—when a particular feature comes out just as you wanted it the first time. Not that common, but it does happen because you're getting better. It's a feedback loop for you too.

In our case of integrating agents into coding Nozbe, we have to be much more careful because we have GDPR, user data, and user tasks and projects. You're in the same boat—you have customer data and you don't want leaks. We're small companies; if we have a catastrophe like that, we're basically done, so we have to really watch out. But with my own iOS apps, I'm the only user, and the worst thing that can happen is a crash on my iPhone, so I have my own sandbox where I can be more liberal. Even though I don't know how to code in Swift, I can say, "Look, you implemented this, but it feels kind of laggy. Can you review it and speed it up?" And it replies, "You're right, Michael, genius idea! I found these bottlenecks and I'm fixing them." Because it has no emotional attachment to its previous work, it just coldly reviews the code.

That's the thing: it's accessible. You don't know how to code, yet you're building viable, hardened applications. My background is different—I started around age 15 writing computer games, even had a couple published in WHSmith. That was all done in 6502 8-bit assembler; you had to know hex, it was hardcore. Then I went to university to learn high-level programming like C and Visual Basic. If AI makes you superhuman, then in the hands of people with decades of programming experience, the prompting is that much sharper because we can see a bit further ahead.

You don't always have to know everything, as long as you know there is a danger—for example, data residency. With all these companies now being built on AI by people who don't really know coding, there's a risk of data drift: they don't really know where the code is running or where the data lives. That's something people need to sharpen up on. The coding part is straightforward, and it's easy to prompt your AI to handle different data residencies, but you have to know to ask. Especially as we're rolling out in the States—we can't get involved in state-by-state privacy laws, so in our terms and conditions, users adding content to Wi-Fi login pages have to set the correct data residency. Knowing that needed to be an option gives you foresight.

For people just starting out, though, it's not scary and not that difficult. You can start today, and it's never too late to catch up. People write to me on social media asking, "Maybe I'm just too late, maybe it's already too advanced for me?" No, it's not! What I like about both of us is our curiosity. When I was preparing my app for testing through TestFlight, I didn't know how it worked—my team handles that for Nozbe. So I just asked the agent, "What am I supposed to do here? Can you do it?" And it said, "Yeah, I can do most of it—you just click here, here, and here, then let me know." It explained the process, prepared the handoff documents, and took care of it.

The same goes for syncing versions of documents. Again, we'll link in the show notes to the rules I use to work with my agent through Nozbe—they are ready for you to use today, with a step-by-step README guide. And in a month or two, when you're curious if Michael improved the rules, you just ask your AI: "Check if the rules on GitHub were updated and compare them to ours." It will tell you what changed and even flag conflicts if your preferences differ from mine. We have to shift our mindset from "Oh darn, now I have to figure this out" to "I don't figure it out—I ask."

That's the whole essence of what we're seeing: taking away all the busywork so you can get to your goal quicker. You can focus on your outcomes without getting bogged down in the "how." I don't know if you follow Dean Graziosi, but in a presentation I saw, he drew two points: where you are now, and where you want to be, connected by a line called "How." Most people get stuck on that line. What AI does is let you focus on where you want to be, plus the X-factor he drew above the line—a heart, representing your passion and your "why."

If you don't have to worry about the "how," you don't get distracted by writing documentation, test plans, or unit testing—which you and I always do, of course!

Of course! Over my 20-year career at Nozbe, I've worked with dozens of programmers. I counted how many were religious about doing unit testing: zero! They all intended to, they thought it was good practice, but they never did it. AI has no problem with that—it just does the unit testing.

Back in the day as an analyst-developer, turning the requirements document into a test plan and finding novel ways to test hard-to-reach code with test harnesses was tough. Now it's all handled for you.

One of my iOS apps is a clipboard manager—the one missing from iOS—and I built it with Apple Shortcuts support and even an Apple Watch companion app. Recently I added a new feature, and it blew me away: it was build number 90, and all 180 automated tests passed! 180 tests for my small personal iOS app, plus browser screenshots!

And there's something else I've noticed: when developing now, people don't even say, "Oh, this is so much faster." It isn't even up for discussion anymore—everyone knows it's light speed. For example, in October we did two and a half years' worth of work at Fydelia. By the end of the year, since starting in April, we'll have done the equivalent of all nine years of Fydelia's lifetime in eight months. It's unimaginable. I liken it to watching Formula One—nobody watching F1 says, "Wow, those cars are much faster than bicycles!" People who love cycling will still cycle, while F1 is a completely different paradigm.

Which makes me think about the hobbyist side of it too. Let's take a quick break to hear a word from a Nozbe user, and then let's dive deeper into the role of programmers and the human aspect of AI.

*(Sponsor break – Nozbe user testimonial)*

And we're back! Before we continue, a quick reminder: today we're launching our Black Friday promo at Nozbe. If you buy or renew an annual plan from today (November 18) through the end of Thanksgiving weekend, you not only save two months compared to monthly billing, you get **3 extra months of Nozbe for free**—15 months total. It's the best time of year to wrap up your annual review, plan the next year, and have a trusted productivity system for the whole year ahead. No promo code needed, and it comes with our full money-back guarantee.

So, Steve: programmers—aren't they scared about losing their jobs now?

This is really where the conversation leads. Looking at the psychological aspect: coming from a hardcore programming background, I always called myself an entrepreneur, but I realized I'm actually more of a builder and an artist—I love bringing things into existence. So what happens to programmers who love the art of coding when AI does it at light speed?

On one hand, it frees the programmer to become the orchestrator and craftsman, seeing the whole product come to life. On the other hand, maybe they'll miss the hands-on craft—like telling a painter, "Don't paint that landscape, we can just print it." For my team, the passion is focused on building a great product quickly.

At the same time, I think we're returning to the era of the hobbyist. Before, you'd finish work completely burned out, unable to look at a screen, just wanting to collapse in front of the TV. Now, when I finish work, I still have mental bandwidth left. I have a soldering iron in my office and I just got a 3D printer. Maybe we're getting our hobbies back because the workday isn't draining us in the same way.

For me, I've always wanted to do something with my hands after staring at a screen all day. That's why I also bought a 3D printer—organizing things with systems like Gridfinity and OpenGrid (which I'll write about on my blog soon).

Secondly, I think programmers gain a great programming buddy in the AI agent. Just like unit testing, everyone always said pair programming was super efficient, but in reality it was boring for the person just watching someone else type. When your pair programmer is an AI agent that's much faster than you, you can spitball ideas, test approaches, and do tag-team development.

That's an interesting concept—though you have to know when to hit the brakes, because the AI is never going to stop generating ideas and it will drain you.

For me, the agent is like an eager puppy: "What can I do for you now? Can I fix this? Can I move forward?" When I landed in Gdynia for our Nozbe reunion, I wrote a very sophisticated prompt to my agent: "Hey." Just from "Hey," it checked our Nozbe project and replied, "Michael, I saw a new task with an obvious bug—can I fix it?" And my second prompt was: "Do it."

What I've also discovered, though, is that because the agent codes so quickly, I now spend much more time thinking about design and UX: "Do I really want this button here, or should we move it to the right?" Before, you would have been happy the button was on the page and worked after three hours of coding! Now there's a focus shift toward quality and refinement. In about six months, our company will reach the point where our design ideas need more time to mature than the actual coding takes.

That reminds me of Agile Scrum and Kanban—moving tickets from To-Do to Doing to Done across a board. Recently we found you can't even keep up with moving tickets across the board anymore—you'd need a desk fan to blow them across! I'd love to ask Jeff Sutherland, who wrote the book on Scrum, what happens now.

And what about product ownership? If features get built so fast, a single product owner runs dry of ideas because developers come back in minutes asking, "That's done, what's next?" You end up with distributed product ownership.

Exactly—programmers will have to become product owners for their parts of the product, iterating quickly without plugging in the head product manager at every single step. The product manager becomes more of a product strategist handling governance and overall vision.

From a productivity standpoint, we might already be at the point where we work fewer hours—maybe six hours instead of eight—because after six hours of rapid decisions, decision fatigue kicks in. You've accomplished a week's worth of work in a day, and your brain is simply done. As a CEO, I can swivel my chair and switch context to sales strategy or accounting, which gives some relief.

Our team is completely distributed—a true "no office" setup. We have our 11 o'clock stand-up, and beyond that I don't track where people are. One of our team members, AJ, takes a walk every afternoon before sunset for an hour or two to decompress, because having a hyperactive AI that never stops can feel oppressive if you don't hit the brakes.

It's like that scene in *Interstellar* on the planet near the black hole where every hour is seven years on Earth! When you stop working with AI, you almost get anxiety thinking, "If I stop for an hour, I'm losing days or weeks of progress." When I was traveling with my family, I caught myself pausing during a walk to send a task to my agent, thinking, "Oh man, the AI is idle right now!" You have to learn to be okay with that, put the phone away, and be present with your family or disconnect on weekends. Though on Fridays I do sometimes think about what I can load up for Monday morning!

So besides using ChatGPT or Claude as a chat, I have my agent set up in Nozbe, and I hear you're running 18 Grok agents on your phone and Mac! How do you incorporate them into your workflow?

They each have their own set of skills. The very first one I created was a "secretary" bot. As a CEO doing business development, I'd go to bed worrying: "Did I reply to everyone? Is someone waiting on me?" So I connected a bot to my Gmail to scan the last few weeks for open-ended threads. At first I had to train it—it flagged an out-of-office auto-reply as an unanswered email—but it learned. Now every morning I get a briefing: "So-and-so has been waiting 23 hours," or "Don't forget you promised to follow up with Jim today." That took a huge weight off my shoulders.

Then I created a couple of developer bots so that on dog walks, I can use voice mode to talk through feature ideas—not critical-path architecture, which stays with the team, but client-facing improvements I spot from the crow's nest. To keep everything in check, I created a "manager" bot that filters my instructions down to the other bots with strict no-go rules: never deploy to production, run specific tests before moving from dev to staging, and so on.

I'm also writing a book called *Relentless*, hopefully finished by the end of the year. In quiet moments, I'll hop on a voice call with my book bot and say, "Interview me for 15 minutes about chapter six." Now we're up to 18 bots for social media, sales outreach, and more. My tip is: start small, solve your single biggest pain point first, and let it snowball. How did you get started?

After my AI course in September, I felt a bit overwhelmed, and right around that time our developers finished the remote and local MCP connectors for Nozbe. So I decided to dogfood our own feature and test working with an agent right inside Nozbe. Right now I'm building a marketing and social media agent to help our team. Full disclosure: our MCP integration still has a few rough edges we're ironing out, and testing it myself helps us spot what needs fixing.

We also built a separate prototyping environment where we can assign a task to a coding agent, it builds the feature, and a build server compiles a test version of Nozbe right away. Often the initial code is a bit of "AI slop" that needs refinement, so putting it into real production with strict security guardrails takes much more work, which is in progress right now. But being able to tap on a real prototype on your phone instead of staring at static mockups is a game changer.

And nobody likes cleaning up files or reorganizing folders—agents love doing that! You can ask an agent to tidy up your Mac desktop or Google Drive. Just make sure you set guardrails so it doesn't do what happened to me once when an improper `rsync` command wiped my hard drive!

We had similar rules for deployments—agents don't get access to staging or production, we use SSH certificates instead of passwords, and live deployments are scheduled pulls rather than pushes. That lets everyone play like kids in a candy store in their own isolated dev environments without breaking production. We're also setting up AI-assisted first-pass code reviews so human developers can stay in that "crow's nest" view—spotting architectural issues while the AI catches cross-system dependencies, like reminding me to hide an unsupported feature from the Apple Watch menu.

So the three pillars we've uncovered for working with AI are **curiosity**, **patience**, and **trust**—learning to let go of day-to-day micromanagement while keeping the overall product vision.

We've been talking for an hour—we just lost a year of AI progress in an hour of chatting! To wrap up: try different tools like Claude Code, OpenAI Codex, or Grok, and keep your rules in Markdown files so they're portable across models. (I tried teaching my daughter Markdown instead of Word, and she told me, "Dad, I'm not a geek like you!")

Steve, where can people find you and Fydelia?

It's Fydelia with a Y—[Fydelia.com](https://fydelia.com) (that's why we don't let AI do our voiceovers, because it pronounces it "Fidelia"!). Look me up on LinkedIn — @steve-Parkinson and follow us on Instagram at `@fydeliawifi`. Always happy to connect, especially if you're in Spain or London.

Thanks again, Steve! Don't forget our Black Friday offer at Nozbe—get 3 extra months free on an annual plan as we gear up to celebrate 20 years of Nozbe!

Thank you for being an amazing listener of the No Office podcast. Every other Wednesday we meet to talk about productivity and hybrid lifestyle, because we believe that work is not a place to go, it's a thing to do. As a special gift only to No Office podcast listeners, when you sign up for Nozbe using the link `nozbe.com/podcast`, you'll get $30 in credits to upgrade to Nozbe Premium. Nozbe is free for up to three active projects and three people on your team, so start today and claim your free bonus credits. See you in the next episode!

Creators and Guests

Michael Sliwinski
Host
Michael Sliwinski
Leading @Nozbe #productivity app | Writing #NoOffice book on #iPadOnly | Blog: https://t.co/vRZY2YrzsE | Husband & father of 3. 🐘 Find me at https://t.co/hHsFpUHwle
Stop Using AI Just as an Assistant - Make It Better Than You - Steve Parkinson
Broadcast by