In partnership with
✨Welcome back, Entrepreneurs
⏱ 3-min read
Last issue, we covered who owns the distribution channel. This issue, somebody upgraded the engine underneath it. Let's get into it.
⚡ OpenAI Just Rewired the Brain of AI Training
OpenAI partnered with AMD, Broadcom, Intel, Microsoft, and NVIDIA to build MRC — a new networking protocol that spreads GPU data transfers across hundreds of paths at once, cutting training disruptions to near zero.
Image source: OpenAI
It's already running inside OpenAI's largest supercomputers and has trained multiple frontier models. This week they released it publicly through the Open Compute Project.
Why it matters for builders: the models you build on are about to get faster, more reliable, and cheaper to run at scale not because of a new architecture, but because the pipes just got a major upgrade.
🔥This week in AI
Image source: Microsoft
🔹Copilot Cowork: The Agent Layer Just Shipped
Microsoft transformed Copilot into a coworker, now available on iOS and Android with workflow templates and plugins for Power BI, Dynamics 365, and more. Assign tasks, and it completes them in the background.You assign the task. It runs in the background. You come back to finished work. If your product lives inside Microsoft 365, your users are about to ask why they still need to do it manually.
🔹NVIDIA + ServiceNow shipped the enterprise version of the same idea
At ServiceNow Knowledge 2026, two companies launched Project Arc, an autonomous desktop agent for knowledge workers, alongside the consumer-grade Cowork. Built on NVIDIA's OpenShell, Project Arc offers a regulated, audit-trailed alternative for accessing local files and running tasks under enterprise governance.
🔹Anthropic went vertical with finance
Claude now ships pre-built agents for credit underwriting, KYC, and reconciliation, plus connectors to LSEG, FactSet, S&P Global, and Morningstar. The "horizontal AI vs vertical SaaS" debate is over. The labs are going vertical themselves.
🔹 Google's Gemma 4 just got 3x faster — open weights
A new technique cuts Gemma 4 response time by up to 3x with no quality loss. It runs on a regular GPU. If your AI bill is breaking the budget, this is the open-source escape ramp — same story as IBM Granite last week, but from Google and faster.
🔹Perplexity went premium for health
NEJM, BMJ Journals, and BMJ Best Practice are now built into Perplexity for medical questions. The "AI search isn't reliable for health" critique just got expensive to make. If you build in a regulated space, paid trust is the new moat.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
📝Anthropic Just Taught Claude To Dream
Anthropic shipped three capabilities for Claude Managed Agents this week.
Dreaming lets agents review past sessions and update their own memory — Harvey saw completion rates jump 6x.
Outcomes lets agents grade their own work against your rubric until it passes — Wisedocs cut review time in half.
Multiagent Orchestration lets a lead agent split jobs across specialist subagents in parallel — Netflix uses it across thousands of apps.
Image source: Claude
The three things agents couldn't do — learn, self-check, divide labor,all just got fixed.
📝 NotebookLM Mind Maps just got a glow-up
Google just rolled out three upgrades to NotebookLM Mind Maps and if you've been sleeping on this tool, now's the time to wake up.
Customization — Steer your map with specific prompts so it matches your thinking, not a default template
Organization — Rename and share maps instantly, turning solo research into a collaboration artifact
Navigation — Silky smooth transitions between nodes make exploring complex topics actually enjoyable
Here's why this matters for founders: a mind map replaces a 40-page document scan. You see the structure before reading the words. Custom prompts mean the map fits your mental model.
The full tutorial — how to prompt the map, the 4 best use cases, and the share flow founders are using for investor prep is on Medium.
📊 Jensen vs Dario on the AI jobs question
Nvidia's Jensen Huang pushed back on Anthropic's Dario Amodei's claim about entry-level job losses pointing to 500,000+ AI jobs created. McKinsey identified new roles emerging from the disruption. Both are probably right. Both are talking about different timelines.
The real question isn't who wins the debate. It's whether your business is building for the jobs that survive or the ones that don't.
$170K in hackathon prizes are up for grabs if you want to put your answer to the test.
Want it every Sunday?
Renewals stop being a fire drill.
Most churn blindsides the CSM in renewal week. Champion left. Usage dropped. NPS slid months ago.
A colleague in Slack watches the signals around the clock. Your CSMs catch every risk months before renewal.
11,000+ teams use Viktor daily. SOC 2 certified. Your data never trains models.
🧬Your body knew before you did.
Your HRV is the signal. Your wearable is already tracking it. Most founders never look.
This week's Wednesday edition: the heart metric predicting burnout before your brain catches up, the Stanford breathing technique that beat meditation in a head-to-head trial, and the 60-second reset you can do between meetings today.
Want it weekly?
