AI agents just crossed a line. For the first time, a non-technical team can direct AI to build production software, automate multi-platform publishing, replace agency retainers, and deploy those same systems across a portfolio of businesses. Not as a demo. Not as a proof of concept. In production, today.
This really began in December, with the launch of Claude Opus 4.5. Combined with Claude Code, smaller, more nimble, AI-first teams can finally leverage AI to re-invent their operations and improve results.
This changes who wins in online business. Large companies built scale the only way they could: by hiring hundreds of people. That model is now a liability. Smaller, AI-native teams are competing at a fraction of the cost, with capabilities that structurally increase over time. The incumbents that don’t adapt will get disrupted. The ones that do will look very different.
Onfolio is built for exactly this moment.
Garry Kasparov observed something surprising about “freestyle” chess tournaments, where human-machine teams compete against each other: “Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.”
The winning variable wasn’t the strongest player or the most powerful computer. It was the quality of the process connecting them.
That’s what we’re building: an AI-native holding company. Not a company that “uses AI” the way every public company now claims to. A company where AI is embedded in how we operate, how we build, and how we create value across every business we acquire. The process is the product. We build it once, refine it continuously, and deploy it across every business in the portfolio. Each new acquisition plugs into a system that’s already been refined through months of real use.
Here’s where we stand.
The Strategy: Three Years In
In June 2023, I published a four-pillar AI strategy. An actual operating plan for how AI would transform how we acquire and operate businesses.
The pillars: 1) Buy businesses and use AI to improve them. 2) Build AI tools for existing businesses. 3) Use free tools for customer acquisition. 4) Buy or build standalone AI tools.
In our 2024 annual letter, I had to be honest: “adoption has actually been fairly thin.” Outside of content and image generators, the practical tools just weren’t there yet. The thesis was right, but the tools were about two years behind where they needed to be.
Sure, you could dive deep, spend a few weeks learning n8n and other technical tools and build automations, but it typically meant immersing yourself in a still-technical world and then figuring out how to apply it to your business. It was a full-time job, and nobody in our company had time to do that and do their main job.
Then a couple of months ago I discovered how much better Claude Code had become. Hat-Tip to Gael Breton and Mark Webster at Authority Hacker, who run an AI Accelerator program that got me up to speed fast. Within days I went from “AI is theoretically useful” to “AI is rapidly rebuilding our infrastructure.”
Why This Time Is Different
I don’t write code. Never have. My job is strategy, deals, and figuring out how to make a small team punch above its weight. But here’s the high-level takeaway: Claude Code is not just for coding. It is arguably even more beneficial for marketers and digital business owners than for pure developers. Its best use case is not coding, it is building and automating.
For a good visual representation of how dramatic the shift has become, take a look at my GitHub activity the past 12 months:
[GitHub activity screenshot]
This is a common pattern I’m seeing across other “Claudepilled” marketers and their teams. People are shipping new ways of working, and it’s not just about coding apps, it’s about re-building workflows.
I want to explain why this matters more than every other “we’re using AI” announcement you’ve seen from a public company.
The shift from chatbots to agents is what changed everything. Previous AI tools required you to manage the output. You’d ask it to write something, check if it was right, fix the mistakes, and try again. Agents manage themselves. They write the code, check if it works, fix it if it doesn’t, and keep going until the job is done. That’s why a non-coder can now build production systems.
Previous AI tools are also friction-heavy to improve. ChatGPT and Claude’s web app have memory features, but updating them is a manual process. You have to go back into the project settings, edit the system prompt, upload new documents. In practice, most people don’t bother unless it’s a big project or important task. Plus, the output never really gets meaningfully better over time.
Claude Code is a different category. It operates inside your actual work environment. It reads your files, writes code, runs scripts, calls APIs. If a newsletter draft uses language that’s too promotional, I tell it to fix the draft and update its style guide so it doesn’t happen again. Five seconds. Next session, the fix is permanent. No going back into settings. No editing prompts. The system just gets better as you use it.
The result is that the system compounds. After a few weeks of regular use, it knows my voice, my compliance constraints, my publishing workflow, and the specific mistakes I don’t want repeated. That compounding is the real unlock. Not smarter AI. Less friction between having a good idea and actually executing on it.
Why a Holding Company Is the Ideal AI Vehicle
Most companies adopting AI get one shot at the ROI. They integrate it into one business, one workflow, one team. If it works, great. If not, sunk cost.
A holding company is different. Every system we build for one business can be deployed across the entire portfolio. The content engine I built for Onfolio’s investor communications applies to Proofread Anywhere’s student marketing, Vital Reaction’s e-commerce content, our agencies’ client work. Each new deployment takes hours, not months.
And every new acquisition we make plugs into these systems from day one. That means the effective cost of operating each incremental acquisition decreases. The more businesses we own, the more valuable the AI infrastructure becomes. That’s a flywheel that most companies can’t replicate.
Ben Thompson recently wrote that the rise of agents means you don’t need mass adoption for AI to have massive economic impact. You just need a small number of people with initiative directing agents across multiple workloads. A holding company is the ideal structure for exactly that. One team, multiple agents, deployed across every business in the portfolio.
For knowledge-work businesses like the ones we operate, this is transformative.
Pillar 1: Buy Businesses and Use AI to Improve Them
Score: Active and delivering results.
For this pillar, it starts with improving our own portfolio. Once we do this, we can build a playbook for acquiring companies that haven’t adopted AI yet, and we have a recipe for instant margin improvement and revenue growth.
There are two very concrete ways AI is changing our businesses in dollar terms. First, work that previously required an agency or a hire can now be done internally. Second, work that we didn’t have the budget or bandwidth for before is now possible. Both improve margins. Both compound over time.
Here’s where this pillar stands across the portfolio.
A Complete Investor Content Engine
This was the first thing I built. We went from “Dom posts on LinkedIn sometimes” to a full automated content system that gets better over time.
The pipeline: I share raw thoughts, rough notes, maybe voice memos from a call. From a single session, Claude can generate anywhere from 5 to 10 different themes, and each theme can have multiple article ideas. When ready, it drafts five formats simultaneously: blog post, newsletter, X thread, LinkedIn post, and video script. Each one adapted for the platform, not just copy-pasted. Editing is easy within VSCode, and I can give feedback, highlight parts I don’t like, recommend rewrites, or point out inaccuracies. Claude makes a note of it to avoid the same mistake in the future and updates every format at once.
Once I approve a post, it schedules everything automatically. The Typefully API handles X and LinkedIn. The WordPress REST API handles the blog. The ActiveCampaign API handles the newsletter to our subscriber list. Images are generated automatically for each platform with different sizes and pillar-coded colors.
What actually shipped: I went from writing 2—3 articles per year to having 20+ pieces of content scheduled across four platforms. The quality is better, not worse. Claude also handled email capture forms, automated welcome sequences, and a content calendar that rotates through seven pillars and tracks three investor archetypes.
The honest part: this would have been a marketing hire’s first quarter of work. I built the entire system in an afternoon.
The best part is, this system is portable to every business in our portfolio. Some of them haven’t been putting much content out at all, others have been using custom GPTs. This system beats both hands down.
Something I didn’t expect: since turning the system on in early February, investor sentiment has improved dramatically. I’ve seen a large increase in engagement, overwhelmingly positive, from both shareholders and non-shareholders. Multiple shareholders told me they increased their position off the back of this content. Even the negative replies are constructive. And I feed all of it back into the content system to generate ideas for future posts or clarifications.
The system doesn’t just publish. It listens and adapts. For a micro-cap with no analyst coverage, that might be the most valuable thing AI built.
Eastern Standard
Eastern Standard is a branding and web agency, and now uses Claude for the vast majority of its development work. Previously Claude was used to quickly gut-check code and review it. Now it’s good enough to do the bulk of the heavy lifting. Are Eastern Standard going to fire all their coders? No, but those same developers are now more efficient and can handle more work, faster, and without dealing with the monotonous tasks that kill productivity.
Additionally, a key team member at ES moved on at the end of 2025, and instead of struggling to find a replacement, the ES management found they could train the number two person in that role to handle the gap using Claude and her existing expertise. That saves payroll, but more importantly improves efficiency.
Over the rest of 2026, ES is expanding Claude into marketing, design, and sales.
An Ad Spend Intelligence Dashboard
This was our COO, Adam. He doesn’t code either.
He built a dashboard that pulls our Hyros ad tracking data for one of our businesses and does something most analytics tools don’t: it tells you what to do next. Is ad spend efficient? If not, why? If it is, can we scale? If so, where?
The next iteration will auto-generate the ads themselves based on what’s performing. That’s $5,000/mo in agency costs or marketing hires avoided for this particular business, plus insights we weren’t getting before.
RevenueZen
RevenueZen is actively integrating AI workflows into client delivery. A lot of the work RevenueZen performs can be built as Claude skills, and we’re already seeing what this does for their margins and capacity.
Will clients simply use AI themselves? In some cases, maybe. But agents still require agency. Someone has to know what to build, what questions to ask, and how all the pieces fit together. Even Adam was only able to build his Hyros dashboard because he knew what he wanted and what data he needed. Most people hire agencies not because they lack the resources to do the work themselves, but because they don’t have the subject matter expertise to direct the work. That doesn’t change with AI. If anything, it becomes more true: the gap between “having AI tools” and “knowing what to do with them” is where agencies create value.
The honest nuance: some agencies are absolutely going to have to reinvent themselves. The agency model was already under pressure, and AI is going to change it fast. We’re going to see a demand for more output for less cost, simply because clients know it’s possible. That said, there will still be demand for premium services, and agencies that evolve and embrace this will do well. Others will go obsolete.
A Cultural Shift, Not Just a Top-Down Initiative
Perhaps the most encouraging sign: one of our business leaders has independently started rebuilding his entire tech stack with AI and is making real progress. He recently told his team, “We need to start thinking of ourselves as an AI company first.” That wasn’t a directive from me. That’s a leader inside the portfolio seeing what’s possible and reorganizing his business around it. When adoption happens bottom-up, not just top-down, that’s when you know it’s real.
Pillar 2: Build AI Tools for Existing Businesses
Score: First product nearly live.
One of our businesses, Proofread Anywhere, is evolving into a broader freelancing platform. For a long time I’ve wanted to build a membership element, but was limited by the fact nobody on the team had the time, the business didn’t have the cash, and it would be a mammoth project with unknown ROI.
Last month, I built the bones and some of the meat in a weekend. Built entirely with Claude Code. This is a membership area with a style checker combining rule-based parsing with AI review, a job aggregator pulling from six sources, a portfolio builder with multiple fully-designed templates, a rate calculator, a proposal generator, a message board, a 30-day career starter area, and full Stripe integration for subscription billing.
Previously, it would’ve cost us tens of thousands of dollars and taken months. It just wasn’t worth the attempt. But now, we directed the architecture and made the product decisions. Claude Code wrote the code. The result is a custom platform that does exactly what our students need, not what some SaaS template allows.
And even if it doesn’t make us much money, we can build, launch, test, and move on in weeks. Claude makes it possible to build things that weren’t cost-effective before. You can build niche tools for small audiences now, because the cost is essentially a weekend of focused work.
Pillar 3: Use Free Tools for Customer Acquisition
Score: In development.
Definitely doable, and this is where most people are focusing their AI efforts. We’re prioritizing other pillars first, but we’ve learned how quick and easy this will be once we shift focus.
Pillar 4: Buy or Build Standalone AI Products
Score: Emerging. First product in development.
This is where the AI-native holding company model creates a unique advantage. Every tool we build internally for one business becomes a potential product for others.
The content system I built for Onfolio’s investor communications is becoming a standalone service offering. The ad intelligence dashboard has the same potential. We have more AI-powered services in development that we’ll announce when they’re ready.
One of the benefits of a holding company is that we get to see what works across multiple properties and adapt it. With Claude, that adaptation takes an hour or two of teaching it about each business and then it builds the system.
What This Means for the Business
Let me be direct about what AI adoption changes for a holding company like ours.
It lowers the cost of building things. When a membership platform goes from $30,000—50,000 to a weekend of focused work, and an ad dashboard goes from $5,000/month to something your COO builds himself, the math on what’s worth building changes completely. Things that weren’t viable at $50,000 are absolutely viable at near-zero marginal cost.
It makes small teams more capable. Eastern Standard absorbed the work of a departed team member without a replacement hire. Across the portfolio, work that took four hours takes one. For a holding company running a growing portfolio with limited headcount, this is the difference between scaling and stalling.
It compounds across the portfolio. Every system built for one business can be deployed to the others. Each new deployment takes hours, not months. And every new acquisition gets plugged into these systems from day one.
It opens new revenue streams. The tools we’re building internally are becoming products. That’s a path from “operator that uses AI” to “company that builds and sells AI-powered products.” For a micro-cap, that’s a meaningful evolution.
To put real numbers on what we’ve seen so far: tens of thousands in development costs avoided, thousands in monthly writing and agency costs reduced, hours of analysis replaced by minutes, and dozens of hours of monotonous uploading, connecting, and formatting time saved.
Why This Matters Beyond Our Portfolio
There’s a broader shift happening that’s relevant to how we think about acquisitions and operations.
We’re watching the early stages of what some people are calling a “SaaSpocalypse.” Large, bloated software companies and agencies built their businesses the only way that was possible before: by hiring hundreds of people. That was the cost of scale. Now, dramatically smaller teams built with AI from the beginning are competing with those incumbents at a fraction of the cost, with more capabilities, and those capabilities are structurally increasing over time.
This creates opportunity on both sides. As acquirers, businesses that haven’t adapted to AI will trade at lower multiples because buyers see the risk. That’s good for our deal pipeline. As operators, the businesses we already own get more competitive against larger players because AI lets a five-person team deliver what used to require twenty.
The companies that will thrive are the ones that figure this out now, not when it’s obvious to everyone. The ones that wait will find themselves competing against leaner, faster competitors who built AI into their operations from the start.
For a holding company, every business we acquire is another chance to deploy this playbook. And every business that hasn’t adopted AI is a potential acquisition target trading at a discount to what it could be worth under our operating model.
The Honest Part
The 2023 strategy assumed AI tools were further along than they were. The tools caught up to the thesis about two years later than expected.
I also underestimated how much human investment AI adoption requires. This is not plug-and-play. The systems work because we spent time building instruction files, testing outputs, correcting mistakes, and refining processes.
AI is a force multiplier, not a replacement for knowing what you’re doing. That actually makes it more of a moat. The businesses and teams that invest in learning how to use it well will pull ahead of those waiting for it to become push-button simple.
Where We Go From Here
All four pillars are now active. Three are delivering measurable results, and the fourth is emerging with our first AI-powered product in development.
The pattern is the same everywhere. AI handles the production. Humans handle the judgment. The businesses that figure this out first operate like companies three times their size.
For Onfolio, the infrastructure is in place, and the compounding effect across the portfolio means each new acquisition benefits from everything we’ve already built. That’s what makes an AI-native holding company different from a company that simply uses AI: the advantage is structural, and it grows with every business we add.
I’ll share updated numbers on how this is affecting operations in future quarterly updates via the newsletter. If you want to follow along, subscribe at onfolio.com/subscribe.
Disclaimer: This post reflects my personal observations and opinions about AI adoption at Onfolio Holdings. It is not financial advice and should not be construed as a solicitation to buy or sell any security. For complete financial information about Onfolio Holdings (ONFO), refer to our SEC filings at sec.gov.