TELEPORT

Run AI agents without exposing your infrastructure

Agents running on static secrets are a liability. Overpermissioned IAM roles are worse. Neither should be the default when running agents against production infrastructure.

Teleport Beams provisions each agent in an isolated Firecracker VM with a short-lived identity certificate injected at startup. Access internal services through Teleport’s application proxy using identity-based, role-driven controls, with no long-lived API keys. Services stay private by default, with access granted specifically via short-lived credentials.

Every access event is logged immutably, with session replay available with enterprise trial.

Provision agent’s access in seconds, with infrastructure that’s isolated, and auditable.

Explore Teleport Beams

_{*This is sponsored}

THROUGH THE VALLEY

1/ Figure is moving humanoid robots from prototype to production

Figure AI has gone from producing one humanoid robot per day to one every hour in under 120 days. Inside its BotQ factory in California, it’s now building its third-generation system, Figure 03 (see the reveal), across 150+ workstations with dedicated assembly lines and layered quality checks.

They’ve produced 350+ robots so far and are targeting up to 50,000 per year, which not long ago sounded unrealistic but now looks actively within reach.

Why this matters:

The bottleneck in robotics is shifting. It used to be about not having enough real-world machines to learn from. Now it’s about whether those machines actually hold up in the wild, doing repetitive work over time without quietly breaking down.

If reliability is there, progress compounds because every deployed robot feeds the next iteration. If it isn’t, scale just exposes the problem faster.

2/ AI spots pancreatic cancer years before doctors

At the Mayo Clinic, a model called REDMOD was tested on nearly 2,000 historical CT scans that had already been reviewed and marked as normal. It still identified early signs of pancreatic cancer in 73% of cases, sometimes up to three years before diagnosis. Around the two-year mark, it detected roughly three times more cases than radiologists.

Why this matters

Pancreatic cancer isn’t hard to confirm once it’s obvious. The problem is that by then it’s usually too late to treat effectively, which makes timing everything.

Most attempts to fix this rely on new tests or additional screening, which rarely scale because they add cost, friction, and more decisions into an already complex system.

This approach pulls earlier signals from scans that already exist, moving detection forward without changing behavior. That’s why it has a real chance of becoming standard, not just studied.

3/ $130B in AI spending, and supply still can’t meet demand

Microsoft, Alphabet, Amazon, and Meta together spent around $130 billion in a single quarter, largely driven by AI infrastructure, and still reported the same issue. Demand is ahead of supply.

Why this matters

These companies aren’t struggling to sell AI, they’re struggling to deliver it. Billions in demand are already sitting in backlogs, waiting on infrastructure that hasn’t been built yet.

That shifts the competition toward capacity. The advantage now comes from how quickly you can deploy compute at scale, not just how good your models are.

TRENDING TOOLS

WHAT I'M CONSUMING

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at barsee@aivalley.ai or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email barsee@aivalley.ai with the subject “AI Valley Ads”.