The future and potential of AI for driving innovation is never standing still. Opportunities seem limitless – and finding ways to focus and apply that power requires constant rethinking.
This year’s TechCrunch Sessions: AI at UC Berkeley pulled together the people building, funding, and regulating the next generation of AI. As expected, the conversation wasn’t just about tech; it was about the implications, the power shifts, and the practical realities of what comes next.
RealityMine’s Chris Havemann (CEO) and Luke Biggins (CCO) were on hand -- here are their five biggest takeaways from the event, and why we think they matter.
A standout moment came during the fireside chat with Jared Kaplan, Co-Founder and Chief Science Officer at Anthropic. Prompted by Maxwell Zeff’s question around recent reports of Anthropic bots visiting Reddit over 100,000 times, the conversation turned to data ethics.
Whether LLMs obtain fair use training data based on clear publisher consent (and economic exchange), or scrape publicly available information where consent isn’t explicitly withheld, the tension is clear: we’re still building the future with foundations we don’t fully understand or regulate.
And as those foundations harden, the need for clearer guardrails becomes impossible to ignore — not just technically, but ethically and legally. Startups working in AI or data spaces should pay attention: your approach to data provenance could become a differentiator, or a liability.
The key issue? How can we ensure Large Language Models are trained on consented, privacy-safe data — with ‘fair’ returns for publishers - and who gets to define those standards?
Kisson Lin (Tanka) introduced a provocative idea: what if AI isn’t just assisting you, but actually co-founding with you?
It’s easy to write this off as a thought experiment – but with human teams often being imperfect, slow to align, and often pulled in different directions, AI, by contrast, offers a version of constant availability, rationality, and scale. If agentic systems become intelligent collaborators (not just task-completers), we could see entirely new models of business formation and leadership emerge.
It’s a glimpse into a potential future where AI isn’t just a tool — it’s a strategic partner.
The question isn’t whether AI will change how we work — but whether it will also change who we work with.
One of the most pragmatic takeaways came from a session on startup positioning. If you're a start-up, your strength lies in agility, so the advice was clear: don’t race the LLMs.
Instead, carve out space where foundational model providers can’t easily go — whether that’s due to regulation, data access, niche workflows, or customer intimacy.
Legacy systems are a disadvantage — until they’re not. Small, unencumbered teams can innovate faster, test more boldly, and deliver more tailored solutions. The goal? Don’t aim for the ‘last mile.’ Build something for the long haul, rooted in deep customer understanding.
There’s a temptation to chase the next big leap — but sometimes, value lives in the margins. One of the most grounded takeaways was the impact of AI that simply helps users do things more efficiently. Simplicity scales — and solving everyday inefficiencies can drive meaningful impact at enterprise level.
Even enterprises like Toyota aren’t chasing novelty for novelty’s sake. They want tools that integrate into daily operations and actually work. That means solutions that make teams faster, smarter, or more consistent — not just flashier.
Startups that can identify those friction points and solve them well will likely find more traction than those chasing moonshots with no clear use case.
Iliana Quinonez (Google Cloud) shared what might be the clearest forecast of the day: we’re heading into a tidal wave of AI agent adoption. Today, around 10% of enterprises are experimenting with agents. Within 2–3 years, that number is expected to hit 80%.
These aren’t static bots, like chatbots or copilots. We’re talking about collaborative, learning-driven systems that will increasingly operate with autonomy and adapt to complex environments.
The real shift? From generic, one-size-fits-all AI to agents trained on your data, solving your problems, and embedded into your ecosystem. The winners will be the ones who can train and deploy these systems responsibly, without drowning in complexity or compliance risk.
We’re moving from general-purpose AI to deeply embedded, purpose-built systems. The companies that win won’t be the ones with the loudest tech — they’ll be the ones who combine insight, ethics, and usability in ways that actually scale.
