💡 The $60M Bet on Fixing AI's Biggest Blind Spot

Sponsored by

☕ Morning! 💡 Your Weekly 5-Minutes of Caffeine and Tech Clarity

Featured Tool 📊 

Run all your IT from one platform, with AI doing the legwork

This one is interesting.

Atera pulls RMM, patching, and ticketing into one platform, and their AI Copilot works the problem alongside you: triaging issues, generating scripts, pulling insights from your live setup and more. You stay in control, it makes the work move.

30-day free trial, too.

Sign up for free →

Quick Hits 🎯

• 🚀 China's Moonshot AI shipped Kimi K3, now the largest open-source model ever released and it's close enough to top U.S. systems that "open" stopped meaning "behind"

• 🧠 Anthropic and Blackstone are betting the next trillion-dollar AI business isn't a model at all it's getting AI to actually work inside a company's existing systems

• 🏁 The smartest people in the industry now think the real competition quietly moved off the frontier leaderboard and onto the deployment stack

• 💸 Enterprises are buying AI compute faster than their finance teams can figure out what it's actually costing them

• 🖥 IBM's mainframe business, long considered untouchable, is showing real cracks as AI-native rivals eat into modernization budgets

• 🔐 A startup called Oak just raised $60M to fix the identity mess AI agents are creating across enterprise systems like impersonation and data leakage, but for bots

• ✈️ AI travel agency Fora just crossed unicorn status, proof that full-stack personalization is becoming a real business moat, not a demo trick

• 🏥 Bunkerhill Health is showing what it looks like when AI agents move patient outcomes, not just paperwork

🎁 + 2 other stories you might find useful

Our Partner 🎉 

The best voice models, now across all channels

Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.

ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.

Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.

The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.

Start building free

The Big Picture 🖼️

💡 Open Models Just Became a Governance Weapon.

Moonshot AI's Kimi K3 is the largest open-source model ever released, and it's landing close enough to top U.S. systems that "open" no longer means "behind."

That changes who gets to compete: regional players and cost-conscious enterprises can now build on a frontier-grade base without paying frontier-grade licensing fees.

The pressure shifts to proprietary labs to win on safety, efficiency, and compliance instead of raw scale. The record-setting release that's rewriting who counts as a frontier lab is worth understanding whether you write code or just approve the budget for it.

The takeaway: model size stopped being the moat. Distribution and trust are the new ones.

💡 The Money Moved From Models to Implementation.

Anthropic and Blackstone are structuring their next big bet around a simple idea: the model isn't the product anymore, the workflow around it is.

Enterprises don't want a smarter chatbot, they want data integration, governance, and measurable ROI wrapped into something that plugs into their CRM and ERP without six months of custom engineering.

Why the two firms think the next trillion dollars gets made in the plumbing, not the model tells you where the hiring, the funding, and the partnerships are actually headed.

The takeaway: if you're evaluating AI vendors on model quality alone, you're grading the wrong test.

💡 The Scoreboard Everyone's Watching Is the Wrong One.

Benchmark leaderboards make for good headlines, but the people actually deploying AI at scale care about something less exciting: can it run in a regulated environment, on infrastructure they control, with licensing terms their legal team will sign off on.

Open models are winning on exactly those grounds. Why insiders think the real contest already left the frontier is the piece to read before your next "which model should we standardize on" meeting.

The takeaway: the smartest teams are optimizing for deployability, not for leaderboard bragging rights.

💡 Nobody's Compute Bill Matches Their Compute Plan.

Enterprises are provisioning AI infrastructure faster than their own teams can measure what it's costing them, a gap that's becoming its own category of operational risk.

The vendors who win the next 18 months won't be the ones with the fastest chips. They'll be the ones who can show a company exactly where its money goes.

The cost blind spot most procurement teams don't know they have explains why "we'll figure out the bill later" is becoming a boardroom liability.

The takeaway: if your AI budget conversation is still vibes-based, you're already behind the teams treating it like FinOps.

💡 The Mainframe Moat Has a Leak.

IBM's legacy business has survived decades of "the mainframe is dead" predictions, but AI-accelerated, cloud-native workloads are finally putting real pressure on the model.

Modernization isn't optional anymore, it's the price of staying relevant to a generation of engineers who've never touched a terminal emulator. What's actually cracking in IBM's oldest cash cow is a useful case study in what happens when a legacy advantage meets an AI-native expectation.

The takeaway: governance and interoperability now matter as much as raw performance, even for companies that have coasted on lock-in for thirty years.

💡 Identity Is the New Security Perimeter.

AI agents are quietly creating an identity problem nobody built for: bots impersonating other bots, credentials leaking between systems that were never designed to talk to each other.

Oak just raised $60M to build the identity rails underneath that mess, verifiable attestations for agents, not just humans. Why $60M just went toward fixing a problem most security teams haven't named yet is a preview of the next category of enterprise security spend.

The takeaway: agent identity is about to become as standard a line item as SSO used to be.

💡 Personalization Just Became a Defensible Business, Not Just a Feature.

Fora's jump to unicorn status is a signal that end-to-end AI personalization. Data, recommendations, pricing, and booking working as one system, can be a genuine competitive moat rather than a nice demo.

The catch is that personalization this deep needs real auditing, or it drifts into decisions nobody can explain. What Fora's valuation says about where AI-native products are actually winning is worth a read whether you're building consumer AI or just trying to understand why it's suddenly working.

The takeaway: the winners aren't building better chatbots. They're building better decisions, end to end.

💡 The Highest Stakes Are Where Trust Gets Tested First.

Bunkerhill Health's partnership shows what disciplined AI deployment looks like when the downside of a mistake is a patient outcome, not a bad recommendation.

That kind of pressure forces a level of rigor around data provenance and regulatory compliance that most enterprise AI programs still haven't reached. How AI agents are earning a role in patient care without cutting corners on oversight is a preview of the governance bar the rest of the industry will eventually be held to.

The takeaway: healthcare AI isn't the exception to enterprise AI governance. It's the preview.

💡 Onboarding Friction Is the Silent Killer of Enterprise AI Programs.

Sequoia's partnership with Sable is built around a less glamorous but more important problem: getting enterprise users into AI ecosystems without a six-month rollout.

Lower switching costs mean faster policy enforcement and less risk sitting unmanaged across a patchwork of vendors.

What closing the "diffusion gap" actually means for how fast AI gets adopted connects directly to the identity and governance threads running through everything else this week, Glean's approach to reimagining work AI makes the same point from a different angle: governance has to be built into the workflow from day one, not bolted on after launch.

The takeaway: the companies winning enterprise AI right now aren't the ones with the best model. They're the ones who made deployment boring, predictable, and auditable.

Trending Tools 📈

PostHog/posthog (+146 ⭐ per day, 🐍 Python) Link

• All-in-one product analytics, session replay, feature flags, and AI observability in one platform

• Helps teams: diagnose problems and ship fixes faster without stitching together five different tools

codecrafters-io/build-your-own-x (+549 ⭐ per day, 🖹 Markdown) Link

• A hands-on guide to rebuilding major technologies from scratch

• Helps teams: level up engineers fast by teaching system internals instead of just APIs

sindresorhus/awesome (+461 ⭐ per day, no language) Link

• A curated list of standout tools and resources across nearly every domain

• Helps teams: shortcut the research phase when evaluating new libraries or approaches

SigNoz/signoz (+204 ⭐ per day, 🟦 TypeScript) Link

• Open-source, OpenTelemetry-native observability with logs, metrics, traces, and AI-assisted insights

• Helps teams: catch latency and reliability issues across both apps and AI agents in one place

ripienaar/free-for-dev (+134 ⭐ per day, 🧩 HTML) Link

• A running list of free tiers across SaaS, PaaS, and IaaS

• Helps teams: keep dev and test environments running without burning budget

coollabsio/coolify (+106 ⭐ per day, 🧪 PHP) Link

• A self-hosted alternative to Vercel, Heroku, and Netlify

• Helps teams: deploy full-stack apps on their own infrastructure with a familiar workflow

Giskard-AI/giskard-oss (+59 ⭐ per day, 🐢 Python) Link

• An open-source testing and evaluation library built specifically for LLM agents

• Helps teams: catch agent failures before customers do, with standardized benchmarks

josephmisiti/awesome-machine-learning (+48 ⭐ per day, 🐍 Python) Link

• A curated hub of ML frameworks, libraries, and tools

• Helps teams: evaluate new frameworks quickly instead of rediscovering them one GitHub search at a time

Tech Trend of The Week 📊 

🔍 "Kimi K3" and "open source AI model" searches climbed sharply this week

The release of the largest open-source model ever built doesn't just move benchmark charts - it moves search behavior. Developers, procurement teams, and curious professionals all started searching for the same thing within hours of the announcement: what can actually run this, and what does it cost to self-host versus rent from a frontier lab.

The signal: interest in open-weight alternatives isn't a niche developer habit anymore. It's becoming a mainstream budget conversation, and the companies with clear answers on cost and control are the ones capturing that attention.

Our Partner 🎉 

Introducing The First Agentic CRM

Get revenue agents, workflows, and automations across every stage of your motion. Access customer data in real time through Attio's web app, MCP, API, and SDK.

Then Ask Attio anything about your business and get instant answers.

It's the CRM that runs the work behind every win.

Try Attio Now

Hit Reply And Tell Me 💬 

What's the most impressive tech you've seen recently that actually works?

I read every single reply

Go where your customers actually are 📍

Most of the internet copies Reddit.

Blogs, SEO pages, AI models, product reviews. They all pull from the same well.

Reddit is the 3rd biggest site on earth. And it trains the bots your users ask for help.

Odd Angles Media runs Reddit campaigns for over 45 brands each month.

Luckily, we’ve convinced them to give away all of their strategies in a (free) ebook ⬇️⬇️

Grab The Free Ebook (Free) →

Reach the People Who Sign the Checks 💰️ 

40,000+ CTOs and engineering leaders. 96% US-based. 80%+ corporate emails.

They evaluate vendors. They shortlist solutions. They buy.

Advertise with us on Paved → (discount applied)