3 Things CIOs Know That Every AI Founder Should Understand – Issue #24
Spotlight: 3 Things CIOs Know That Every AI Founder Should Understand
This week’s Spotlight is: 3 Things CIOs Know That Every AI Founder Should Understand.
AI is not being blocked by technology. It is being blocked by how companies are built.
At a recent Mayfield dinner with CIOs & Tech leaders, three patterns stood out – and what they mean for founders building in AI:
Insight #1: The bottleneck is not the models. It’s the workflows and decision-making layer.
What it means for founders: Don’t build better tools on top of broken systems. The opportunity is to rebuild the workflow itself, not optimize a step in it.
Insight #2: The workforce impact is already here. Some roles will shrink. Others will be redefined. Demand for systems thinkers, architects, and domain experts is rising quickly.
What it means for founders: You are not just building software. You are reshaping how work gets done and who does it. The winners will build for the new org chart, not the current one.
Insight #3: ROI is hard to measure directly – but value is showing up as speed, better decisions, and reduced friction.
What it means for founders: Sell outcomes differently. The value of AI is not just cost savings. It’s decision velocity and intelligence amplification. We are moving from ROI → Return on Intelligence.
The companies that win won’t experiment more. They will pick one function and rebuild it end-to-end for agents. Not AI layered on workflows. But workflows are rebuilt for AI.
This was clearly evident in the signals this week. Salesforce launched 30+ new Slack AI features that execute multi-step workflows across enterprise tools, Microsoft introduced its own in-house AI models to control more of the stack, and Cursor rolled out collaborative multi-agent systems for coding workflows.
Full Weekend Edition below. 👇
Signals Shaping the Future of AI:
Infrastructure
CoreWeave raises $8.5 billion chip-backed loan to expand AI cloud capacity. The deal, the largest of its kind, shows GPU infrastructure is now financeable at massive scale, reinforcing CoreWeave’s position in the AI cloud market. Click here
NVIDIA invests $2 billion in Marvell to advance silicon photonics for AI infrastructure. The partnership focuses on improving data transfer speeds inside AI clusters and scaling high-bandwidth connectivity between GPUs. Click here
U.S. power equipment shortages delay AI data center buildouts. Limited transformer and grid capacity are forcing reliance on imports, exposing a critical constraint in scaling AI infrastructure. Click here
Enterprise
Microsoft launches its first in-house AI models to reduce reliance on OpenAI. The models are designed to support its Copilot and Azure stack as the company builds deeper control over its AI infrastructure. Click here
Cursor launches a new AI agent experience to take on Claude Code and Codex. The platform enables multiple AI agents to collaborate on tasks such as writing, debugging, and executing code. Click here
Anthropic’s Claude Code roadmap leak reveals new persistent agent features. The “Kairos” update includes memory, background execution, and autonomous workflows for long-running tasks. Click here
Salesforce launches 30+ new Slack AI features, including desktop automation. The update enables Slack to execute multi-step workflows across enterprise tools beyond messaging. Click here
Capital Flows
Anthropic acquires biotech startup Coefficient Bio for ~$400 million. The move signals expansion beyond software into AI-driven life sciences and drug discovery capabilities. Click here
Supabase is reportedly in talks to double its valuation to $10 billion. The database startup is seeing strong demand from developers building AI-native applications, reflecting growing infrastructure demand beyond model providers. Click here
Mistral raises $830 million in debt to build European AI data centers. The financing will fund NVIDIA-powered infrastructure as the company expands regional compute capacity across Europe. Click here
Research
Google releases Gemma 4, an open Apache 2.0 model built for reasoning and agents. The new model family targets agentic workflows and raises pressure on proprietary models by expanding high-performance open-weight capabilities. Click here
Arcee AI open-sources Trinity-Large-Thinking, a 399B MoE model. The release adds a frontier-scale open model under Apache 2.0, enabling full customization and narrowing the gap with closed systems. Click here
Google warns quantum computing threat to encryption may be closer than expected. Researchers estimate elliptic-curve cryptography could be broken with ~20x fewer resources, accelerating security concerns. Click here
Claude Code source code leaked via npm misconfiguration. The incident exposed internal roadmap details before Anthropic issued takedowns, underscoring growing risks around AI tooling supply chains. Click here
Policy
California mandates AI safety standards for state vendors. Governor Newsom signed an executive order requiring AI companies contracting with California to meet new safety, transparency, and privacy guardrails. Click here
EU bans AI-generated content in official government communications. Core EU institutions prohibit the use of fully AI-generated images and videos to enforce authenticity in public-facing materials. Click here
UK opens antitrust probe into Microsoft over AI bundling concerns. Regulators are assessing whether its software unit should receive “strategic market status,” which could trigger new restrictions. Click here
AI lobbying group plans $100+ million political spend in the U.S. Innovation Council Action is funding pro-AI candidates in one of the first major efforts to influence AI regulation at scale. Click here
Global AI Strategy
Microsoft commits $10 billion to AI infrastructure in Japan and plans to train 1 million engineers. The investment, alongside partners SoftBank and Sakura Internet, expands data center capacity while scaling AI talent across the country. Click here
Australia partners with Anthropic to build AI policy and research infrastructure. The initiative combines safety evaluations, adoption tracking, and local research to strengthen national capabilities and position Australia in the global AI ecosystem. Click here
Chinese GPU makers capture 41% of China’s AI server market. Domestic chipmakers are rapidly gaining share, accelerating China’s push to reduce reliance on Nvidia and build a sovereign AI hardware stack. Click here
South Korea chip exports surge 151% YoY to record levels. The spike reflects accelerating global demand for AI hardware, with memory chips emerging as a key driver of the compute cycle. Click here
Talent Signals
Each week, we spotlight key roles tied to the themes shaping this week’s AI headlines, connecting talent to the companies driving the news.
Sycamore builds an enterprise platform for deploying and orchestrating AI agents with built-in governance, security, and lifecycle management. The company recently raised $65 million to develop what it calls an operating system for autonomous AI in the enterprise. As organizations move from pilots to fully deployed agent systems, Sycamore is expanding its engineering and product teams. Open roles are listed on its careers page. Click here
GimletLabs builds infrastructure that routes AI workloads across different types of compute hardware to optimize inference cost and performance. As inference becomes a key constraint in AI deployment, Gimlet is growing its engineering and platform teams. Open roles are listed on its careers page. Click here
Upscale is redesigning the networking layer for AI, building open, high-performance network infrastructure to connect GPUs at scale. As traditional cloud networking struggles to keep up with the speed, scale, and latency demands of modern AI workloads, Upscale is creating the backbone needed to support next-generation training and inference systems. Open roles are listed on its careers page. Click here
You can see all the opportunities at Mayfield-backed AI companies here, and across the broader ecosystem here.
Social Signals
The most important conversations in AI are unfolding across social media, where top voices are shaping the next wave of signals and strategy. Here are some of the top social signals and their takes from the past week.
Dan Martell (Click here) — “For thousands of years, every organization has run on the same logic the Roman Army invented. The whole structure exists to solve a bandwidth problem. Jack’s argument is simple: AI solves it better. Block built a world model, a continuously updated picture of everything happening across the company. When the world model carries the information, you don’t need the layers.” In a post viewed 658K+ times reacting to Jack Dorsey’s essay, Martell argues that AI could fundamentally reshape org design by removing layers built for information flow. Instead of hierarchy managing communication, real-time systems handle coordination, shifting leverage toward smaller teams, faster decisions, and outcome ownership.
Gergely Orosz (Click here) — “Anthropic accidentally leaked the source code of Claude Code. Repos sharing it are being taken down. But one repo rewrote the code in Python, so it violates no copyright and cannot be taken down.” In a post viewed 2.1M+ times, Orosz highlights how quickly proprietary AI systems can be reverse-engineered and redistributed once exposed. The episode underscores the fragility of closed-source advantages and how open implementations can rapidly emerge, even when original code is removed.
Evan Luthra (Click here) —“Matthew Gallagher started Medvi from his house in Los Angeles. AI wrote the code. AI made the website. AI made the ads. AI handled customer service. First full year $401 million in sales, and this year on track for $1.8 billion.” In a post viewed 1.3M+ times, Luthra highlights how Gallagher built Medvi with just two people, himself and his brother, reaching scale typically associated with thousands of employees. The example reflects how AI is compressing company-building, shifting the advantage toward speed, execution, and the ability to orchestrate tools rather than build large teams.
