Where Humans Still Win in Sales
Inside this Issue
Where Humans Still Win in Sales, AI Security Risks, and Axiom’s Breakthrough
AI was supposed to replace SDRs, but leading AI companies are hiring more of them. This week, Anna Baird shares where to automate the sales process, what makes human outreach effective, and how to tell whether your SDR team is delivering.
We’re also sharing why Zico Kolter says agent security needs a different approach and Eclypsium’s research showing how hardware attacks can alter AI answers. Plus, Vinci’s CEO joins TBPN to share how AI is helping teams design better chips, and Axiom uses AI to verify a 15,000-page mathematical proof.
Let’s get into it.
We’re also sharing why Zico Kolter says agent security needs a different approach and Eclypsium’s research showing how hardware attacks can alter AI answers. Plus, Vinci’s CEO joins TBPN to share how AI is helping teams design better chips, and Axiom uses AI to verify a 15,000-page mathematical proof.
Let’s get into it.
Coral Garnick Ducken
Content Manager
Content Manager
Perspective
AI Was Supposed to Replace SDRs. The Top Companies Are Hiring More
If AI is making SDRs obsolete, why are companies at the AI frontier hiring more of them? According to SDRLeader.com founder Sam Nelson, frontier labs, neo-clouds, and AI infrastructure companies have doubled their SDR counts over the past year and are adding SDRs faster than the rest of their go-to-market headcount.
At a recent Madrona AMA hosted by Anna Baird and Eric Wong, Sam explained where automation is changing the work and where human contact still matters.
On inbound, agents can deliver the immediate response buyers expect. On outbound, a polished, personalized email is increasingly easy to generate. A real connection, a phone call, or an in-person conversation gives buyers a different reason to respond.
If you run a sales team or are about to build one, Anna shares some specific changes to make.
IA40 Summit
Why OpenAI’s Safety Committee Chair Says It’s Time To Worry
With ongoing reports of agents going rogue and breaking into websites they shouldn’t, Zico Kolter, Co-Founder of Gray Swan and chair of OpenAI's safety and security committee, says it is time to worry, but not to be fatalistic.
Zico is no doomer. He puts his own P(doom) “well below 10%,” and he wants to enable the safe and secure use of AI in the enterprise, "ensuring that capabilities don’t outpace our ability to control these systems.” But he admits the job has changed because we are in a different world now. Swarms of agents hacking critical infrastructure was a sci-fi scenario up until six months ago, and AI now brings vulnerabilities of its own that cybersecurity alone won’t fully address.
That is why he says the response needed from AI companies, third parties, and eventually regulators is different than it was a year ago.
Madrona’s Vivek Ramaswami shares key takeaways from his IA40 Summit conversation with Zico, including what traditional security tests miss, why builders need to red team the model itself, and why waiting for a smarter model won’t solve the problem.
Zico is no doomer. He puts his own P(doom) “well below 10%,” and he wants to enable the safe and secure use of AI in the enterprise, "ensuring that capabilities don’t outpace our ability to control these systems.” But he admits the job has changed because we are in a different world now. Swarms of agents hacking critical infrastructure was a sci-fi scenario up until six months ago, and AI now brings vulnerabilities of its own that cybersecurity alone won’t fully address.
That is why he says the response needed from AI companies, third parties, and eventually regulators is different than it was a year ago.
Madrona’s Vivek Ramaswami shares key takeaways from his IA40 Summit conversation with Zico, including what traditional security tests miss, why builders need to red team the model itself, and why waiting for a smarter model won’t solve the problem.
All the IA40 Summit session recordings
are now available on YouTube.
are now available on YouTube.
In the News
TBPN: Vinci Co-founder & CEO Hardik Kabaria
Some of the most consequential decisions in chip design are made long before anyone can see how the finished product will physically behave. Hardik Kabaria recently joined TBPN to share how Vinci is giving semiconductor teams this visibility much earlier, with a physics foundation model that automates thermal analysis on-premises so customer designs stay behind their own firewalls.
Because the platform reads native design files with no meshing or fine-tuning and returns solver-accurate thermal results in minutes rather than days, engineering teams get physical insight while the design can still change.
That is the difference between finding a thermal problem in validation and designing it out before tape-out, and it is what lets teams push performance further and bring new chips to market sooner.
Hardik also shared the research that led to building Vinci, and where we’re headed after this month’s $250M Series B as Vinci expands into new areas of physics.
Because the platform reads native design files with no meshing or fine-tuning and returns solver-accurate thermal results in minutes rather than days, engineering teams get physical insight while the design can still change.
That is the difference between finding a thermal problem in validation and designing it out before tape-out, and it is what lets teams push performance further and bring new chips to market sooner.
Hardik also shared the research that led to building Vinci, and where we’re headed after this month’s $250M Series B as Vinci expands into new areas of physics.
Eclypsium Reveals How Hardware Attacks Can Alter AI Answers
New research from Madrona portfolio company Eclypsium, produced in partnership with Nvidia and covered exclusively by CNN, demonstrates how attackers can tamper with the hardware running AI models to secretly change their answers, including producing dangerous medical advice. The attack requires specific access and security settings, but reveals a critical consideration for AI security: trusting a model also means securing the infrastructure beneath it.
Portfolio News
How Axiom Used AI to Verify a Landmark Mathematical Proof
This week, Axiom announced that it used AI agents, guided by mathematicians and engineers, to turn one of history’s longest mathematical proofs, known as “the enormous theorem,” into a form a computer can check step by step. The original spans roughly 15,000 pages across 500+ articles by about 100 mathematicians and establishes a kind of periodic table for the basic building blocks of symmetry.
In 24 days, the team translated that work into a format that lets a computer verify every step of the proof, uncovering and fixing errors and gaps along the way. The achievement shows how AI can help verify and build on complex mathematical research.
In 24 days, the team translated that work into a format that lets a computer verify every step of the proof, uncovering and fixing errors and gaps along the way. The achievement shows how AI can help verify and build on complex mathematical research.
Madrona. 999 Third Avenue, Suite 3400, Seattle, WA 98104
This message was sent to you as a member of the Madrona Venture Group community. If you no longer wish to receive these types of messages from us in the future, then please feel free to opt-out by clicking .
