My feelings about AI are unresolved

I’ve been going a little bit crazy trying to reconcile the two dominant ways of thinking about AI. I simply cannot avoid hearing about how the technology promises more than it delivers, is a vehicle companies can use to substitute agents for actual employees, risks deskilling professionals, appropriates other people’s labor without permission of compensation, consumes extraordinary amounts of capital and energy, convincingly lies to you, and distorts or degrades social relationships that are at the core of human progress.

Sounds bad, right? 

But swing the pendulum the other way and we see a wave of curiosity and enthusiasm about what AI helps unlock. Just look at how many have installed OpenClaw despite the potential security risks. Even people like Claire Vo, who were initially skeptical, are now using it to manage their businesses and their lives. Ryan Rumsey showed us how he connected a design system to a Claude skill to generate HTML and Figma prototypes that faithfully aligned to those standards. Joey Banks walked us through the ways the Figma Console MCP can be used to streamline design system workflows. Or from a product perspective, consider Polyend and their Endless effects pedal that leverages AI to help people vibe code their own custom instrument effects. 

Examples of designers using AI to achieve their goals without putting their faith in a content-generating slot machine are plentiful. The technology is not infinitely useful nor is it monolithically ill-suited for all tasks. It’s murky. As Eryk Salvaggio notes in Stochastic Flocks and the Critical Problem of ‘Useful’ AI, “What remains urgently in dispute are the boundaries of utility: what usefulness means, for whom, and under what conditions?” In between the boomers and the doomers, is a cohort of people using AI to make the drudgery go away and clear a space where there is more time to focus on what makes these experiences legitimately meaningful.

Watching this all unfold from the sidelines wasn’t getting me anywhere. Thinking is for doing, so I gave myself a challenge. As someone with no mobile development skills and minimal experience with current AI tools, would I be able to convert a vague idea for a mobile app into a fully functional, SwiftUI-native iOS application and get it listed in the App Store? How would I feel about the process and the outcome? There was a metaphor that kept recurring in my head: would I feel like the character of Ripley in Aliens (played by Sigourney Weaver), operating a Power Loader and amplifying my own capabilities? Or would I feel more like Edgar in Men In Black (played by Vincent D’Onofrio), an alien invader poorly mimicking a human in an AI-powered skin suit? I had to start building so I could find out for myself!

The Product

The seeds of my app idea came from a post by Ronnie Batista where he noted the concept of Shoshin which posits, “In the beginner’s mind there are many ideas; in the expert’s mind there are few.” The algorithms that power our online experiences can narrow our thinking by filtering content that aligns with our preferences. What if there was a light, daily ritual that could encourage us to see the same thing in different ways?

The app attempts to create a space for this ritual by prompting users with a single photograph and inviting them to see the same image from distinct perspectives. When you’re done, the system uses AI to look at your responses and observe the characteristics of what you wrote through sentiment and structural analysis. There are no achievements, streaks, or badges and it does not compare what you write with anyone else who might be doing the same. It aims simply to be a quiet daily ritual that favors presence over progress.

The Process

On an episode of Lenny’s Podcast, Lazar Jovanovic described his vibe coding workflow using Lovable so I started with their custom GPTs that helped generate the scaffolding and set some boundaries for the tool. From there I followed a winding path, with the occasional backtrack, that put me in contact with most of the AI tools at one point or another. Eventually I realized that Lovable could not deliver a native mobile app experience and that it’s suggestion to use CapacitorJS caused more problems down the road. This shift lead me from Lovable to GitHub to Cursor to Xcode with Claude. There was even a brief dalliance with Firefly along the way. I learned many hard lessons at each point, but some stand out more than others

You Win Some…

1.     I could experiment, rip things out, and change things with little fear. Lovable asked me how I wanted to handle authentication and I reflexively chose username/password. With near immediate regret (because no one will want to remember a password for something this casual), I swapped it for a magic link, which broke the Lovable preview. I swapped it again for a one-time password. I should have thought about this a little harder before blurting out my responses, but the penalty for changing my mind was trivial.

2.     I could embrace my utter lack of skill and lean on the agent for guidance. This was especially helpful during the App Store submission process which was…not intuitive.

3.     Designing in the medium you are designing for is powerful. This is not a prototype or a limited concept. There is no script to adhere to. For better or worse, the application works exactly as I intended.

4.     Claude insists the app has an accessibility score of 95/100 and has successfully implemented Dynamic Type and VoiceOver support. It is WCAG AA+ compliant as is the companion website (which got an AIM Score of 9.8 out of 10).

You Lose Some…

1.     The agent lies, convincingly and repeatedly. [Al Jourgensen Voice] Never trust an agent. Instead, try to set up a system that helps an agent check and verify the work that it does. Daniel Roth, editor in chief right here at LinkedIn, talked with Claire Vo about how he uses apps called “Bob the Builder” and “Ray the Reviewer”. He requires Bob to stop work frequently and run plans past Ray. Doing something like this, documenting everything, and working in small increments feels like a good way to manage context windows and make sure your project stays on track.

2.     The agent misdiagnoses issues. For me, a weird bug during authentication was wrecking account creation and sign in. I would leave to get my OTP code and come back to a blank screen. The agent went through 10 or 11 different diagnoses, each time insisting that it had finally figured out the issue, but each time it hadn’t. It ultimately recommended refactoring the whole authentication workflow, creating a compounding tangle of new complexity and failures in the process. Starting a new conversation that I fed the documentation from all our previous attempts identified and fixed the issue in about two minutes, but we spent another 30 undoing all that refactoring work.

3.     The agent flatters and fluffs and praises in ways that I am consciously suspicious of and still susceptible to. Literally everything is a great question. My ideas are refreshing and perfect. Working with me is a pleasure and a gift and we have accomplished so many great things together. It’s disturbing and feels manipulative. When I had finished all my testing and the app was ready for submission to Apple, Claude gave me a vigorous handshake and sent me off with conspicuous fanfare. I felt like a character in A Nightmare on Elm Street having to remind myself over and over: You’re not real!

4.     The project started affecting my sleep. As I got closer to the finish line, I had more trouble falling asleep and staying there. I woke up thinking about prompts or issues that had not gotten adequate attention like security or performance. I speculated about how I might trick the agent into improving things I lack the expertise to provide direction on. It was not AI brain fry, the mental fatigue and cognitive overload some have discussed during heavy AI usage, but I can appreciate how easily that might happen.

Shipping is the Morale Event

I finished v1.0 of Eye•Full and successfully launched it in the App Store. The first round of bug fixes will be submitted soon, and a batch of feature enhancements will be released soon after that. I find myself nodding in agreement with Erika Flowers who said “You are allowed to be furious about AI and still learn to use it. You are allowed to grieve the career you planned and still build the next one. You are allowed to think this is all moving too fast and still move with it.” I am suspicious of AI. Wary, even. But I also have a working app in the App Store and a growing interest in personal software, the small applications tailor-made to solve problems corporate software can’t justify the investment for.

So where did I land on the Ripley/Edgar question? A bit of both, to be honest. I was able to orchestrate things using AI that I could not do on my own. It’s powerful stuff! But I also put myself in situations where I was forced to trust an entity I do not, and should not, have naïve faith in. My confidence in the app, its infrastructure, its reliability, and its maintainability would be much higher if I was able to consult with an actual expert. For Eye•Full the stakes are low, but If I wanted to pursue something more consequential, these gaps would feel like much more of a burden. Perhaps there is a better metaphor for vibe coding with AI in this moment:

AI certainly has a role to play yet, for good or ill. Where are you on the Ripley/Edgar spectrum now? Or, are you more of a Gollum?

RIYL:

Here are a some links to things that I mentioned above:

From Luddite to Shoshin: Holding on to Our Creative Spirits in the Face of AI by Ronnie Batista: https://medium.com/@rjbattleship/from-luddite-to-shoshin-holding-on-to-our-creative-spirits-in-the-face-of-ai-cbe4df73e3eb

The rise of the professional vibe coder (a new AI-era job): https://www.youtube.com/watch?v=0XNkUdzxiZI

Lovable GPTs by Lazar Jovanovic: https://lovable-gpts.lovable.app/

From journalist to app developer using Claude Code from the How I AI podcast: https://www.youtube.com/watch?v=HbWu_eYIHKQ

The MCP Tool That’s Changing How I Use Figma by Joey Banks: https://www.youtube.com/watch?v=lwUCs6ci3Kg

Turning a Design System into a Claude Skill by Ryan Rumsey: https://www.youtube.com/watch?v=rE_H4-AybMc

Stochastic Flocks and the Critical Problem of ‘Useful’ AI by Eryk Salvaggio: https://www.techpolicy.press/stochastic-flocks-and-the-critical-problem-of-useful-ai/

Mutually Assured Construction by Erika Flowers: https://eflowers.substack.com/p/mutually-assured-construction

Here’s more information about Eye•Full: https://eyefull.app/, https://apps.apple.com/us/app/eye-full/id6760787641