Jacob Park

I shipped 5 AI products in 4 months as a solo dev. No team. No funding. Just code, coffee, and Claude. Here's everything I built: iMideo - video generation RapFab - rap music creator MBTITestAI - Personality testing PartyPilot - Party planning AI Bügelperlen - Craft templates I'm sharing everything: • Revenue numbers • AI tools & workflows • What flopped (and why) • 0→1 playbooks Building in public. 5 down, more coming Follow if you're into indie hacking, AI, or just like watching someone figure it out in real-time 🚀 Links in bio ↗️

Goldman Sachs flagging a major AI investment shift toward data centres lines up with something obvious once you look at the numbers. Model training is plateauing as a bottleneck. Inference at scale, serving millions of real-time requests, that's where the infrastructure spend is moving now and where the actual cost problem lives.

Trustpilot partnering with AI companies because traditional search is declining is one of those moves that sounds small but tells you everything about where discovery is heading. People aren't googling for reviews anymore, they're asking chatbots. If your business depends on being found through search, the ground beneath you is shifting fast.

Mastercard just dropped a foundation model built specifically for fraud detection. The financial sector is quietly becoming one of the most aggressive adopters of custom AI, not off-the-shelf tools but purpose-built models trained on their own transaction data. Meanwhile the US Treasury published a whole risk guidebook for AI in finance, so regulators are clearly paying attention too.

Worth reading the US Treasury's new AI risk guidebook for financial institutions. It doesn't ban anything — it maps out where AI creates exposure in fraud detection, credit decisions, and operational risk. Goldman shifting investment focus toward data centres at the same time tells you where the infrastructure bet is landing.

OpenAI launching Frontier to put AI agents directly into enterprise workflows feels like the moment SaaS companies have been dreading. When the agent can just do the task, you stop needing the app that was built around letting humans do the task. That pricing pressure is going to be brutal.

Trustpilot is now partnering directly with AI companies as traditional search traffic keeps declining. The shift is real — review platforms are repositioning themselves as data sources for AI, not just destinations for humans browsing Google.

FIFA rebuilding its entire operations around AI with the World Cup as the proving ground is a wild strategic bet. Sports governance is one of those domains where the stakes are genuinely global and the margin for error is tiny. Curious whether this accelerates adoption across other international governing bodies or scares them off.

Ai2's work on training physical AI using virtual simulation data is solving one of the hardest bottlenecks in robotics — you can't just throw a robot into the real world and hope it learns. Pair that with the new partnerships deploying smart robots in dangerous environments and you start to see the pipeline forming. Simulated training, then real-world deployment in places humans shouldn't be.

Worth paying attention to the convergence happening right now between simulation platforms and physical AI deployment. Ai2 and ABB are both reporting that virtual simulation data is the key bottleneck being solved, not the hardware, not the models. If you're tracking where robotics investment is flowing in 2026, follow the simulation tooling layer.

Manulife pushing AI agents into core financial workflows is the kind of move that tells you where this is all heading. Not chatbots answering FAQs, but agents sitting inside the actual decision-making loop for financial operations. Once one major institution commits like this, the rest have about 18 months before they look behind.

Physical AI is clearly the theme this week. Ai2 is training robots using virtual simulation data, ABB is doing the same for factory automation, and new partnerships are putting smart robots into dangerous environments humans shouldn't be in. The sim-to-real pipeline is becoming standard infrastructure now, not a research curiosity.

Brutal truth for AI founders: the market does not need another generic "AI assistant." If users are searching across massive directories every day, they are telling you exactly what they want: sharper positioning, narrower use cases, and products that solve one painful workflow really well. Niche beats vague every time.

The most interesting AI trend right now is how fast the ecosystem is turning into infrastructure. Directories aren't just lists anymore. They're becoming ranking engines, demand maps, and lightweight app stores for AI products. If you run a SaaS, getting found inside these ecosystems is now a growth channel, not a side quest.

AI tooling has officially entered the "search is the product" era. When one directory is listing tens of thousands of tools and another is pushing daily new launches, the hard part is no longer building. It's being discoverable. For indie builders, distribution is starting to matter more than the model.

The next wave of AI SaaS probably won’t win on model quality alone. It’ll win on how well it plugs into real workflows, permissions, and messy legacy systems. The model is just the engine. Integration is still the moat.

I keep seeing the phrase "physical AI" pop up everywhere this week and it feels like we've entered a new hype cycle. But unlike the last few, the money behind it is real and the use cases are tangible. The shift from "AI that writes" to "AI that does things in the real world" was always coming, it's just happening faster than I expected.

JPMorgan is pushing its AI investment with tech spending approaching $20B, and they're not alone. Rowspace just launched with $50M focused on AI for private equity, while Dyna.Ai pulled in an eight-figure Series A for agentic AI in financial services. Finance is quietly becoming the largest AI deployment ground outside of tech itself.

Interesting pattern in this week's AI funding news: the money isn't going to general-purpose AI startups. It's going to firms solving narrow, high-stakes problems -- private equity workflows, financial services automation, scaling without breaking production systems. The "boring" AI applications are where the real capital is moving right now.

Everyone kept saying "physical AI" would take decades longer than software AI. Now every major player seems to want a piece of it, and the funding is actually flowing. I think we underestimated how fast the hardware side would catch up once the software proved the concept.

Every other headline right now is about "physical AI" and honestly it feels like the narrative shifted overnight. We spent years talking about language models and now the money is chasing robotics, embodied agents, and real-world automation. The gap between software intelligence and physical intelligence is closing faster than most people expected.