WGMaTW: Saying No in the AI Arms Race, Following the Money, and Giving AI Its Own Machine
What Got My Attention This Week — as a Product Leader
Here’s what got my attention this week — a list of interesting things I read and watched, as a product leader.
1/ The Power of Saying No in an AI Gold Rush
“What if the best product decision is saying no to what everyone else is building?”
In an episode of Supra Insider, Marc Baselga and Ben Erez sat down with Alexander Danilowicz, founder and CEO of Magic Patterns.
Magic Patterns is one of my favorite products right now and I appreciate their ability to focus — in their case, it’s on frontend development, prototyping and design systems.
While other AI prototyping tools like Replit, Bolt, and v0 are rushing to add backend capabilities, Magic Patterns is deliberately refusing to do so.
The episode gets into product quality, DNA, risk, feedback, presets, integrations, dogfooding, prompt engineering, and strong leadership relationships.
If you’ve ever wrestled with feature pressure that doesn’t align with your business vision/mission/strategy — even amidst a wild AI gold rush — this one’s worth your time.
Sometimes, the best decision is saying no to what everyone else is building. That’s if you get it right, of course.
2/ Where AI Dollars Are Actually Flowing
ICONIQ Capital released a 2025 State of AI report — surveying ~300 execs (CEO, Head of AI, Eng, Product, CRO, CFO) at companies shipping AI in production.
A few takeaways:
Nearly 80% of AI-native builders are investing in agentic workflows, or autonomous systems designed to take multi-step actions on behalf of users.
70% of companies are building vertical AI apps.
More than one-third (37%) plan to adjust their pricing in the next year.
Teams are using ~3.1 model providers on average — multi-model is cleary the norm.
AI/ML engineers take the longest to hire of any AI-specific role, with an average time-to-fill exceeding 70 days.
Differentiation is happening at the application, workflow, data and model layers.
3/ Give AI Its Own Machine
If we’re serious about AI coding agents, we may need to give them their own machine.
Not a tab or side window, but a dedicated environment. Pushing the limits of delegation and automation, beyond assistance and autocompletion.
That’s what DDH and others have been asking in response to OpenClaw.
“With OpenClaw you're giving AI its own machine, long-term memory, reminders, and persistent execution... It's a sneak peek at a future where everyone has a personal agent assistant, and it's fascinating.”
— David Heinemeier Hansson, 37Signals
The question for product leaders:
What UIs, and constraints are we designing and building for? And what do we need to align to, anticipate, and get ready to reject or rethink? Apps, systems, copilots, agents, best practices, processes, team sizes, business models, etc. It’s all going to look very different by the end of this decade.
“I don’t know what I did.”
— Peter Steinberger, OpenClaw
Bonus: The myth that search is dying
Graphite partnered with Similarweb to analyze ~40,000 of the largest U.S. sites to evaluate the trend in organic search traffic. Their conclusion is that the prevailing narrative that SEO traffic has dramatically declined is false.
“The Myth: Organic traffic from Google to websites is down dramatically (between -25% to -60%). The Truth: Organic traffic from Google to websites is down slightly (-2.5% YoY).”
— Graphite
