Founded & Funded

Who is responsible when an AI agent does something its user never intended? Janel Thamkul, former deputy general counsel at Anthropic, explains why the answer depends on your contracts, your product design, and the records you keep.
In this episode of Founded & Funded, Madrona General Counsel Joanna Black talks with Janel about the legal decisions every AI founder needs to get right, including:
1) Why AI's constrained supply changes how you incentivize go-to-market teams
2) How a product's UI reveals whether a team understands agentic liability
3) The OpenClaw gym class incident and what it shows about open-ended tasks
4) Why copyright damages can be company bankrupting, and how model providers responded
5) Why granular logging is cheap to build early and brutal to retrofit
Liability for AI agents is still an open question, and founders are making product and contract decisions before the answers arrive. Janel gives concrete choices a team can make before shipping.
Transcript/Summary: https://www.madrona.com/what-founders-should-know-about-ai-legal-risk-before-they-ship
00:00 – Introduction
01:16 – What Founders Building on Frontier AI Models Underestimate: Supply Constraints and Sales Incentives
03:06 – How to Tell If an AI Team Understands Agentic Liability by Looking at the UI
05:38 – The OpenClaw Gym Class Incident: When an AI Agent Exploits a Vulnerability
07:56 – Who Is Liable When an AI Agent Fails? Contracts, Copyright, and Indemnification
10:23 – AI Safeguards That Are Cheap Early and Brutal Later: Granular Logging
11:36 – Why the Ecosystem Isn't Ready for Agentic AI: Principals, Trust, and Identity
13:30 – Why AI Startups Should Not Ship Without Real Terms of Service
15:27 – Don't Be the Slowest Zebra: How Early-Stage AI Companies Should Take on Risk
17:11 – Own Your Most Important Relationship: Why No Third Party Should Sit Between You and Your Customer
19:21 – Final Advice for CEOs: Tokenmaxxing, AI Stack Sovereignty, and the ZEBRA Mnemonic

Recorded live at Seattle Tech Week, this episode of Founded & Funded brings together two people who track the AI capital markets from different vantage points:  ⁨Pitchbook⁩  EVP of Research & Market Intelligence Nizar Tarhuni, and Madrona Partner Sabrina Albert.
These two dig into where AI venture capital is actually landing across the model, infrastructure, and application layers, why AI "harnesses" are becoming a winner-takes-most category, and where vertical AI applications may hold an edge that horizontal platforms can't match. They also cover PitchBook's VC Exit Predictor, the new normal for AI seed valuations, and the tranche-financing structures Nizar says are red flags for founders.
And in a live Q&A, they take on a frequently asked question in the AI market: Is this a bubble? Nizar explains why he thinks it's still too early to call, what the data says about the value already being created, and why the next few stages of company growth will tell us much more about which valuations prove durable.
For founders raising capital right now, or operators trying to figure out where durable value is emerging in AI, this conversation offers a data-grounded look at what's happening beneath the headline numbers.
Full Transcript: https://www.madrona.com/is-ai-a-bubble-what-pitchbook-nizar-tarhuni-sees-in-the-data  
Chapters: 
00:00 — Introduction
04:11 — The Rise of AI Harnesses
06:50 — Vertical vs Horizontal AI Applications
07:51 — Vertical AI Use Cases: Legal, Finance, Healthcare
09:25 — Are Vertical AI Companies Defensible vs OpenAI/Anthropic?
11:32 — Durability, Switching Costs & the Exit Predictor
14:44 — Open Source Models & Cost Management
18:58 — Choosing the Right Model for the Task
20:09 — Funding Environment: Seed to Series B
23:04 — Tranche Financing & Founder Red Flags
26:43 — Audience Q&A: Seattle's Role in AI, Bubble Risk & Services as Software

After leaving a long sales career, David Rubenstein spent a year meeting more than 300 founders across 41 countries and turned what he heard into a framework for how companies sell now. In this episode of Founded & Funded, David sits down with Madrona's Anna Baird and Eric Wong to dig into why AI left buyers with more information and less clarity, why deals stall out of fear rather than disinterest, and how founders can cut through.
The conversation covers selling against build-it-yourself AI objections, the return of in-person selling, and why the founder's own brand is more important to drive pipeline.
This was originally an AMA with Madrona portfolio companies, but we wanted to share David's full sales playbook with founders, GTM leaders, and operators for a grounded look at how buying behavior shifted and where to focus when the old tactics stop paying off.

Full Transcript: 
Chapters
(00:00) – Why AI Made Your Buyers Harder to Sell To
(01:30) – How David Met 300 Founders Across 41 Countries in a Year
(03:06) – Why the Old Go-to-Market Math Stopped Working
(04:40) – AI Gave Buyers More Information and Left Them More Confused
(05:27) – The SPRINT Framework for Diagnosing a Stalled Deal
(07:03) – Give the "How" Before the Number So Results Read as Proof
(08:54) – The Question Every Executive Buyer Asks Late in a Deal
(10:16) – 3 Hidden Fears That Kill Deals During Implementation
(12:00) – The Trigger Most Founders Leave Out of Their Customer Profile
(15:24) – Why the First Seven Minutes of a Sales Call Are Wasted
(16:19) – How to Sell Against "I'll Just Build It With Claude Code"
(22:24) – Why AI Sales Prep Gives You the Average, Not the Edge
(25:48) – Why Founder Brand Now Drives Pipeline
(28:50) – Free vs. Paid Pilots and Why Skin in the Game Wins

Joe Beda and Craig McLuckie co-created Kubernetes, the infrastructure standard that became the default for cloud native computing. Now running Stacklok, they're watching enterprises hit the same identity, permissions, and security problems with AI agents that took the container ecosystem years to resolve, and they're building tools to compress that timeline.
In this episode of Founded & Funded, Madrona's Tim Porter sits down with Joe and Craig to talk through what AI adoption actually requires: why MCP is the Docker moment for AI-native applications, how the LLM gateway is becoming a strategic chokepoint for cost, safety, and model flexibility, and why enterprises that don't get the architecture right early will face a familiar trap: vertical integration that looks like productivity and acts like lock-in.
They cover:
Why the developer workflow is the template for knowledge worker AI adoption, and where the analogy breaks down
The mainframe vs. open platform question that will define the AI infrastructure era
Why the knowledge worker transition is harder than it looks — and what has to be built differently before developer-grade AI tooling can scale to the rest of your organization
The governance gap between human accountability and AI behavior, and what enterprises actually need to build to close it
Where to start: MCP controls first, LLM gateway second, and why deploying a platform without staying to close the loop consistently fails
Transcript: https://www.madrona.com/the-best-infrastructure-moment-since-cloud

Chapters:
(0:00) – Introduction
(1:04) – Why the Kubernetes Creators Are the Right People to Read This AI Moment
(2:18) – Joe's Lesson from Cloud Native: Ignore Conventional Wisdom, Except When You Shouldn't
(4:16) – Craig on Enterprises and the Chaos of a New Infrastructure Era
(5:32) – Why Joe Rejoined Craig at Stacklok: The Engineer's Case for Getting Your Hands Dirty
(7:05) – Developers as Agent Orchestrators: How the Knowledge Worker Transition Will Follow
(10:10) – MCP Explained: Craig Sees Docker in 2013 When He Looks at the MCP Spec 1
(7:53) – The Mainframe vs. Open Platform Question That Will Define the AI Era
(20:24) – LLM Lock-In Is the Wrong Worry: The Real Risk Is Left of the Model
(25:19) – Where Enterprises Actually Start: Developer Posture First, Knowledge Workers Second
(29:10) – MCP First, LLM Gateway Second: The Concrete Technical Starting Point
(31:19) – How Stacklok Builds Software Now: Agents, Smaller Teams, the Unrecognizable Developer Profile
(38:07) – The Recruiter Who Started Building Agents: What AI Tools Do to Role Boundaries

There's a theorem being tested about how AI reaches general intelligence. Carina Hong's answer: through mathematics.
Carina is the founder of Axiom, and in less than a year of building, her team's AI has scored a perfect 120/120 on the Putnam mathematical competition — a test where more than 50% of brilliant undergraduates score zero. More concretely, Axiom Prover has reached 98.93% on a Lean software verification benchmark that leading alternatives solve at 11–12%.
In this conversation with Matt McIlwain, Carina explains her central thesis: that math and code are the two pillars of the digital world, and that any AI infrastructure missing a formal verification layer is structurally incomplete. She walks through the history of verified AI research at Google, DeepMind, OpenAI, and Meta, and explains why each effort stalled just as commercial pressure mounted. She describes what makes hardware and software verification the natural first commercial market, and what Axiom discovered when they tested their prover against circuits that industry-standard formal checkers could not verify.
For founders and operators trying to understand what's actually changing in AI capability, and for anyone building in adjacent infrastructure spaces, this is a map of where the frontier is and where it's heading.
Transcript: https://www.madrona.com/agi-needs-formal-reasoning-carina-hong-is-building-it-at-axiom
Chapters: 
(00:00) – Introduction
(02:01) – How to Define AGI Right Now — and Why There Are Two Competing Definitions
(04:17) – Math Is AGI
(06:12) – Math Data Scarcity: Why a Disadvantaged Domain Accelerates Progress
(08:13) – Formal vs. Informal Math: Why AI Researchers Treat This Like a Religion
(13:44) – Google, OpenAI, DeepMind, and Meta All Abandoned Formal Math Research
(21:20) – The Putnam Story: First AI Perfect Score
(28:13) – Hardware Verification as the Commercial Frontier: What Axiom Found Testing Real Circuits
(31:22) – 98.93% vs. 11–12%: What the Benchmark Gap Reveals About Formal Provers
(34:22) – Math and Code as the Two Pillars of the Digital World
(37:00) – Team Building Around a Shared Dream: Recruiting for Mathematical Superintelligence
(38:03) – What Autonomous Proof Generation Looks Like