How We're Thinking About AI at FGO Finance Group

I've been keeping a running collection of the best thinking I've come across on where this is heading. I'll link to it at the end for anyone who wants to go deeper. But first, here's how I'm actually thinking about this as someone running a commercial finance brokerage in Melbourne.

Three predictions worth paying attention to

1. The "one-person team" will become a common operating model within five years.

Anthropic, which officially surpassed a $30 billion annualised revenue run rate this month, has reportedly run significant parts of their growth marketing function with a remarkably small team. That's not an anomaly anymore. When execution gets automated, one person with good judgement and the right systems can run functions that previously required a team of ten. This has already started in marketing, legal review, financial analysis, and content production. Brokerages will feel this too. Certain functions within any brokerage today can already be run more efficiently this way.

2. The premium on judgement is going to accelerate.

Andrew Chen from a16z frames it well: AI compresses the middle 80% of any workflow. What gets more valuable is the first 10% (knowing what to build, what question to ask, what deal to pursue) and the last 10% (quality validation, catching what's actually broken, exercising taste). The people who use AI as a ghostwriter will progressively lose the judgement muscle. The people who use it as a sparring partner will sharpen it. That distinction will define careers over the next decade.

3. "Boring" businesses are the biggest AI tailwind, and ETA is well-positioned for it.

The Sequoia thesis is that every dollar of enterprise software spend has approximately six dollars of professional services spend alongside it. That six dollars is the attack surface for AI. The businesses most likely to benefit are not the ones building AI, they're the ones applying it to established, repeatable workflows where small efficiency gains compound. Services businesses, trades, professional practices. The exact types of businesses that ETA searchers are acquiring, and why I joined a brokerage focused on helping searchers and acquirers with debt advisory services. This is a tailwind worth paying attention to, though it's early days and I wouldn't over-index on it yet.

How I've been experimenting

I'll be honest about where I started. A few months ago I was primarily using AI for chat. Asking questions, getting answers, occasionally drafting an email. Useful but incremental.

What changed was taking experimentation seriously. Not just trying one tool but stress-testing several. Claude for deep work and code. ChatGPT for quick research and brainstorming. Gemini for image editing and generation. I've been following what teams like Nous Research are doing with Hermes, watching how Chinese labs like Qwen are approaching open-source models, keeping up with projects like OpenClaw. The landscape is moving fast and staying across it takes deliberate effort - I shared in a previous LinkedIn post about the premium on time right now. No better time than now to be curious.

The shift that mattered most for me was moving from chat to systems. Building what Andrej Karpathy describes as a "second brain" - a structured knowledge base that AI can read, reference, and build on. I run mine in Obsidian, connected to Claude Code. Every morning I get a brief that synthesises my calendar, inbox, market data, and pipeline into a single view. Content ideas flow from a wiki of concepts so that I can see things from multiple angles, test concepts and iterate with someone who contextually understands what's important and why. Outreach logging, customer intelligence, competitive monitoring - all running as semi-automated workflows (I'm not going to say fully automated, as I have not reached the state where full delegation makes sense).

Jack Dorsey talks about companies needing a "world model" - a compressed understanding of their customers that AI can reason against. That's essentially what I've been trying to build for FGO: a persistent context layer that understands our clients, our market, and our operations well enough to be genuinely useful, not just fast. Speed is great, but context is far more valuable.

What I've actually learned

Don't expect immediate success

It takes more than a couple of iterations to get quality. Don't give up. The third or fourth round is where it starts reflecting how you actually think and what you actually want it to do. Planning is also much better than just winging it. I've found that spending more time designing upfront, thinking through what you want the system to do before you start building, saves enormous time fixing, ripping and replacing after the fact.

Slowly, then suddenly

The compounding is real but slower than I would have expected. A few months into my role at FGO Finance Group and the difference is noticeable. Two weeks in, it wasn't. For SMEs, I predict there is going to be tremendous value unlock (process optimisation that leads directly to margin improvement at the bottom line) for those that know what to change from an operating perspective and what to keep. Not everything is replaceable with AI, and that's the most important part to take away. Some things are far better left untouched, and this is where good judgement and experience is necessary. Don't believe everything it tells you - even if it seems plausible.

Taste will be the valued skill

The last 10% at the start, and the end, is everything. Without taste applied to what AI produces, you get polished, confident slop. I'm still working through this one myself. The models are truly getting really good, but if you lack the initial ability to scope and prompt well, and then you lack good taste once the initial draft comes out, you end up with content that people see straight through as ML generated. The quality bar is set by the human reviewing and refining, not by the model generating.

What this means for how we work at FGO Finance Group

We're a boutique commercial finance brokerage. We don't have large marketing teams, tech teams, or an operations department that helps us manage our customers, run marketing campaigns and make website changes for us. What we have is a system that is beginning to let us operate as if we did.

The true value that I'm working towards isn't in the process and systems. Those are cool (I will admit) and there's value in improving how things work. However, I spend far more time thinking about how we can better service clients, have more regular touchpoints and understand their unique needs. In some cases, we've been able to spot structuring options or lender fits that wouldn't have surfaced in a standard conversation. Other benefits also compound - faster turnaround on deal analysis, more thorough lender matching and superior market intelligence that helps to inform every conversation. The AI doesn't replace the advisory relationship - it makes the relationships we build better by giving us capacity we wouldn't otherwise have.

If you're thinking about how AI applies to your business, whether you're a business owner or an operator considering an acquisition, or an individual navigating AI, I've been collecting the best thinking I've come across over the past few months.

You can find it here: linktr.ee/myaiguide.gabriel