🔄 Regular readers will notice: This is our fifth deep-dive into AI-powered productivity in recent months. If you've been following along with The Multi-Agent Coding Revolution (#98), Carl Vellotti's Claude Code tutorial (#111), McKay Wrigley's Obsidian + Claude Code workflows (#113), Tomer Cohen on full-stack builders (#114), or Maor Shlomo's "vibe coding will kill SaaS" thesis (#115), you'll find familiar territory here - but with a distinctive investor's lens.

SOoooo

Want something completely different? Try these from the archives:

💡 Top tip - Scan the key insights below, then dive into the sections on personal automations and the future of software investing - that's where the most actionable material sits.

Andrew Wilkinson, Co-founder of Tiny, interviewed by Dan Shipper on "AI and I"

Key insights from the full article:

Yes another Claude Code article 🥱 ….but there's a reason why we keep doing them. Different perspectives help you understand how to get the best out of what is clearly the best tool right now

Andrew Wilkinson's enthusiasm is palpable from the opening seconds. After two weeks with Claude Code running Opus 4.5, he describes the experience as having "like 30 free employees working 24/7" for roughly $40 per day. This isn't hyperbole from an AI tourist - Wilkinson runs Tiny, a holding company that acquires and operates businesses, and has been experimenting with AI tools for years.

The difference with Opus 4.5 is qualitative, not just quantitative. "I feel like I have a $100,000 a month payroll of engineers working for me 24/7," he explains. Previous experiments with Replit and other vibe-coding tools always hit walls - bug loops, impaired functionality, things that almost worked but didn't quite. This time is different.

How this compares: Carl Vellotti's Claude Code tutorial (Pod Shots #111) explained how to set up Claude Code and build custom commands - the "beginner to hero" mechanics. Wilkinson's perspective is different: he's showing what becomes possible when a non-technical entrepreneur actually uses it. Where Carl provided the step-by-step setup guide (install with one command, use the CLAUDE file for persistent memory, build slash commands), Wilkinson demonstrates the end state - building a custom email client in a week that would have taken Superhuman "five years in beta."

What makes Anthropic's approach distinctive, according to Wilkinson, is the simultaneous optimisation for both engineering capability and human empathy. "A lot of the models as they get better at programming, they've got worse at other things," he observes. Opus 4.5 maintains that ergonomic, human-like quality whilst delivering code that actually works.

Key Takeaways:

Dan Shipper makes a prediction: the unsung heroes of this AI era will be designers. Wilkinson, whose background is in web design rather than programming, is a perfect case study. "My happy place was headphones on, Photoshop, CSS, front-end," he recalls, "but I would always get so frustrated because from the moment that I produce the stylesheet and the HTML, I can't do the follow-through."

Now he can. The coordination problems that plagued creative work - needing to brief developers, sit through product management meetings, wait for implementation - have dissolved. "I feel like I can really move at the speed of thought," Wilkinson says. "There was a great tweet by a guy who said, 'I finally feel like AI enables me to move at the speed of my ADHD.' And I feel that."

The enterprise view: Tomer Cohen, LinkedIn's CPO, made a similar point in Pod Shots #114 when he predicted that "in a few years you won't have any product managers as you know them - everyone will be what I call full-stack product managers." Where Wilkinson is a designer-turned-builder, Cohen sees the same phenomenon from inside a 25,000-person organisation. The difference? Cohen emphasises the speed of execution changing ("AI agents become the core of how you interact with the product"), while Wilkinson focuses on the elimination of dependency on others.

Shipper notes that his own creative director, Lucas, exemplifies this transformation: "They're like, I've been able to make beautiful things forever, but as soon as I got to coding, it wasn't something I could do." Now designers who understand experience and emotion can take products end-to-end without depending on engineering resources.

Key Takeaways:

Wilkinson receives 200-300 emails daily. Previous solutions required multiple assistants and still left him feeling like Lucy in the chocolate factory, overwhelmed by the conveyor belt. His first attempt at solving this used Lindy to build automations that routed and triaged emails, reducing his load by about 50%.

But the breakthrough came with Claude Code. "I just said to Claude Code, here's my Gmail credentials. I want you to build an email triager. Here's how I want it to work." Within a week, he had a web-based interface that met his exact workflow - something that would have been impossible to build with no-code tools.

The methodical alternative: McKay Wrigley took a different path to a similar destination (Pod Shots #113). Where Wilkinson builds flashy consumer-facing tools, Wrigley demonstrates how to structure knowledge work systematically - building Obsidian-based workflows for research digestion, automatic note tagging, and knowledge synthesis. His tutorial shows 10 increasingly sophisticated workflows that transform simple markdown files into autonomous research assistants. Same tool, same transformative impact, completely different application. Wilkinson's approach is entrepreneurial ("what cool thing can I build?"); Wrigley's is methodical ("how can I systematise my thinking?").

The system ranks emails by priority and importance, provides multiple-choice response options, and for complex emails with multiple questions, offers a Q&A mode where he can answer each question individually before the AI drafts a response. "Anyone who's technical knows how astounding that is," Shipper observes. "Superhuman spent like five years in beta because getting the basics of a good email client is so hard."

Key Takeaways:

Perhaps the most striking example of Wilkinson's AI experimentation is a relationship analysis tool he calls "Deep Personality." The concept started with a conversation with his girlfriend about how useful it would be to have a GPT trained on their relationship dynamics.

He asked Claude what a therapist would want to know to get the best picture of a couple, receiving a list of about 20 clinically validated tests and inventories. Then he built an interface that guides users through these assessments and produces a comprehensive analysis.

"We sat down and we read it out together and we were laughing because it predicted every single fight that we have in our relationship," Wilkinson recounts. The tool identifies personality traits, attachment styles, potential conflict areas, and provides relationship blueprints for couples. He's planning to release it publicly.

The practical application extends to actual conflicts. "We were having a fight and it said, 'Andrew, when you do this, it triggers this thing with his mom, and when he was a kid, he felt that when he did this, he was unlovable.'" When he communicated that insight, "she softened and we're like, 'Oh my god, that's crazy.'"

Key Takeaways:

Wilkinson's wardrobe automation exemplifies how AI can address surprisingly personal problems. Years ago, as a newly single divorcee wanting to improve his appearance, he hired a personal stylist - an expensive process involving shipped clothes that didn't fit and choices he didn't make.

His AI solution is more elegant. He photographed his entire wardrobe and used Claude to convert the images into a CSV spreadsheet. Every morning at 7am, an automation checks the weather in Victoria, generates four outfit recommendations from his actual clothes, and sends them to Nano Banana (an AI image generation tool) to create renderings of him wearing each option.

The result arrives via text: "Wear this watch with this outfit" alongside AI-generated images of him in the clothes. "It has really helped me up my game," he says. He's also created a custom GPT he can query for advice - "What goes with these jeans?" or "How should I style this shirt?" - which responds with specific suggestions like "French tuck that shirt."

Even his facial hair is AI-optimised. "I had ChatGPT do an analysis. I put my face into Pro and said, 'I want you to looks-max me.' And it told me to grow a beard. Now I'm refining the mustache length versus the beard length and dialing it all in."

Key Takeaways:

Wilkinson has built a Lindy agent that records all his meetings and provides an unusual feature: post-meeting psychological analysis. "I have a tendency to really enjoy narcissists," he admits. "I'm drawn to them." Given that psychology is "probably the most important skill in business," he wanted automated detection of potential red flags.

The agent texts him after meetings if it detects concerning patterns. Recently, a contractor who had missed a deadline became defensive and accusatory during their call. "I actually left the call feeling really upset at myself," Wilkinson recalls. "And then I got a text from my Lindy agent about a minute later saying, 'Hey, I just wanted to make you aware this person was using some manipulation tactics. They were gaslighting. They were reframing in this particular way.'"

Multi-agent meets meeting analysis: Kieran Klaassen's deep-dive on multi-agent coding (Pod Shots #98) showed how to orchestrate parallel AI agents for software development. Wilkinson is applying the same architectural pattern to personal productivity - one agent for meetings, another for email triage, a third for style recommendations. The "manager of AI agents" mental model Klaassen described for coding applies equally to life automation.

He emphasises the importance of careful prompting - his system uses a very high bar and only triggers notifications perhaps once every two to three months. The danger, which Shipper highlights, is confirmation bias: AI can reinforce your interpretation of events if you prompt it with your own opinion embedded.

Key Takeaways:

Wilkinson's excitement about building is tempered by serious concerns about software as a business. Tiny has "really slowed down on buying technology companies and software companies," he reveals. His reasoning centres on moats.

"The moat for software used to be it's very expensive to hire programmers and it takes a long time to learn," he explains. "Now it's basically free. And so your moat has to come from something else - a brand, a distribution mechanism, hardware lock-in - because I just don't see them being very good businesses in the long term."

The founder's perspective: Maor Shlomo of Stackfix made this exact argument in Pod Shots #115, predicting "there will be a time where it will be easier to build your own Salesforce-type CRM than to buy a license." Shlomo comes at this as a builder watching his own industry disintegrate; Wilkinson comes at it as an investor watching his portfolio thesis evaporate. Both arrive at the same conclusion: the SaaS model of "build once, sell many" breaks when everyone can build.

He offers a pizza analogy: if a machine could make world-class pizza and anyone could buy it for $10,000, would pizza restaurants disappear? No - but margins would collapse. "Maybe you used to be able to charge 10-15% margins on a slice of pizza, but when the cost goes way down and the quality goes way up, the consumer benefits and the business owner's margin goes to like 1-2%."

The calorie-tracking app boom illustrates this. "A kid vibe-codes a calorie tracker and makes a million dollars a month, and everyone's like, 'That's amazing.' But what happened in the next three months? Everyone who could just copied him."

Key Takeaways:

Wilkinson's strategy for navigating an uncertain future involves owning "toll bridges." "If you have money to invest, I think owning compute or computer power is probably one of the only ways to own a toll bridge to the future."

He shares a specific example: a recent investment in Iren, a Bitcoin mining company that pivoted to AI data centres. "I was able to buy it at a valuation where it was being valued like a Bitcoin miner, not very highly. I invested at a $10 share price and now it's at $35, and recently it was at $70."

His broader concern isn't long-term - "in 20 years we're going to be fine" - but rather the transition period. "What I worry about is another Great Depression where suddenly all these white-collar employees get laid off from previously very lucrative jobs - lawyers, maybe doctors, anything that's knowledge work, programmers." Where do they go? Perhaps blue-collar work, but what happens when a hundred laid-off knowledge workers all start HVAC companies in the same city?

Key Takeaways:

Wilkinson's journey from anxious observer to enthusiastic builder offers a template. A year ago, he called Shipper "freaking out" about whether knowledge work would disappear. Now he's channelling that energy into creation rather than worry.

His advice is practical: educate yourself and prepare, then stop worrying. "In my basement I've got earthquake stuff, a Starlink dish, food. What I've realised about AI is all you can really do is educate yourself and prepare and then not worry about it."

For builders, the message is clear: the tools are finally good enough. "I've never been this excited about work," Wilkinson says. "I wake up at 4am to pee and just get out of bed because I'm so excited to work on this."

For software investors and operators, the implications are more sobering. Unless you have genuine moats beyond the cost of development, the future may look more like pizza restaurants than software companies - technically still businesses, but with fundamentally different economics.

Key Takeaways:

Similar themes explored from different angles:

Want something completely different?

That’s a wrap.

As always, the journey doesn't end here!

Please share and let us know what you liked or want changing! 🚀👋

Alastair 🍽️.