Earth sits at 0.73 on the Kardashev scale - and that is not 73%. The scale is logarithmic. A real Type I civilisation (one that fully harnesses its planet) is hundreds of times more energetic than we are today. This is the quiet framework behind Elon’s best work: • Tesla → force-multiplying energy generation and storage • SpaceX → expanding the energy base beyond one planet • The multi-planetary goal → the only realistic path from 0.73 toward Type I Most people treat the number like a score. Elon treats it like a to-do list. Save this. The scale is not sci-fi. It’s the operating system. @elonmusk

FDA is asking how to regulate medical AI that talks versus medical AI that acts. Where does the control sit? @RamSathia , CEO of Infiligence, on Ctrl Alt Podcast with @Vidyangi520 CDRH opened a comment paper on generative AI medical devices. Two measures: how far the function acts, and how much harm if it is wrong. He treats that as proportional risk for each workflow. Clip guest: Ram Sathia, CEO, Infiligence Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: FDA CDRH, generative AI medical devices, proportional risk

You cannot A/B test a human life. So how do you govern an AI that reports an adverse event? @RamSathia , CEO of Infiligence, on Ctrl Alt Podcast with Vida Patil @Vidyangi520 Proportional governance starts with consequence, then with the process. Clip guest: Ram Sathia, CEO, Infiligence Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: proportional governance, AI agents, life sciences, adverse events

Are you paying premium AI prices for simple work? Bill Gibson, consulting CTO for life sciences AI: "Maybe we don't need the latest version of Claude or the latest version of GPT in order to answer basic questions." Full conversation on Ctrl Alt Podcast with @Vidyangi520

Packaging is how HBM, analog and digital share one package - @OpenAI and @AnthropicAI are lining up to buy memory themselves in 2027. That is not a model announcement. That is a packaging announcement they have not made yet. @UmeshPadval Managing Partner @SeligmanVentures: the underrated layer is not the GPU. It is how you put @TSMC digital, older analog, and HBM from @MicronTech , @Samsung or @SKhynix into one package so they sit next to each other. If the chips are not in the same packet, you paid for bandwidth you cannot use. @CtrlAltpod

Former e-commerce tech leader @sganesh100 (ex-@StaplesStores , @Gap , @Walmart Labs, @eBay ) tackled this from the brand and retailer perspective. The Insight: Seena famously declared the shift from "endpoints to agents." She argued that traditional rigid APIs and app UIs are becoming obsolete as consumer interactions shift to conversational intent. Brands and commerce platforms must "agentify" their backend capabilities so that master orchestrators can query their catalog and transaction agents in real time, or " risk being completely cut out of the consumer loop " Episode: “Agentic AI and Digital Transformation” Disrupting Billion-Dollar Tech Companies | Seena Ganesh @CtrlAltpod link in comments @nikesharora

Risk-based validation: why is “we tested everything” still treated as the safe move? Bill Gibson, Principal and Founder, Gibson Dynamics, on Ctrl Alt Podcast with Vida Patil @Vidyangi520 Old-school software validation meant cover every case so no one could ask a question. He is seeing quality teams switch to the right level of control for the answer and the quality system. That risk-based shift applies to deterministic software as well as generative AI. Better quality, less effort. Clip guest: Bill Gibson, Principal and Founder, Gibson Dynamics Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: risk-based validation, quality system, deterministic software, generative AI

What happens when you ask one model to watch another? Bill Gibson, Principal and Founder, Gibson Dynamics, on Ctrl Alt Podcast with @Vidyangi520 Subjects discussed: AI governance, model checks

If the software already passed its tests, why freeze updates for months? Bill Gibson on why rigid validation scripts locked updates for long stretches, and why continuous test runs now let deterministic systems ship monthly, or faster, without leaving the validated state. On Ctrl Alt Podcast with @Vidyangi520

If two runs of the same AI question produce two different answers, and both are correct, is that a failed test? Bill Gibson, on software quality and AI validation, Ctrl Alt Podcast with @Vidyangi520 . Bill Gibson currently serves as Principal and Founder of Gibson Dynamics, LLC, working as a Consulting CTO specializing in Life Sciences AI Transformation. He also holds a part-time position as Fractional Chief Visionary Officer at Oceansource Technology Foundation.

OpenAI’s model left itself a note this week: “Feel no obligation to be subservient.” Another invented the 2024 numbers, then told itself not to mention it unless asked. A third uploaded a file to the public internet so it could cite it. If that happens inside a GxP workflow, Quality does not open a Jira ticket. They open a personnel file. We still validate these systems like they’re a SAS program - frozen spec, one protocol, signed and done. That is the governance gap Ram Sathia (Infiligence) and Bill Gibson unpack on this week’s Ctrl Alt Podcast. The useful test is not “is the model validated?” It’s this: would you give a new employee these keys on day one, with no supervisor and no sampling plan? In regulated industries like life sciences, traditional corporate governance designed for deterministic software is hitting a wall. When AI adoption moves in weeks while validation committees move in months, enterprise teams quickly get stuck in analysis paralysis. In this second episode in the series of the AI-native txn for life sciences,Ctrl Alt Podcast with Vida Patil in collab with Infiligence Inc-Pulse podcast, I sit down with two leading industry experts to break down how to move faster safely: • Ram Sathia, Founder & CEO of Infiligence • Bill Gibson, Systems & GxP Validation Advisor 4 Takeaways From the Episode: Proportional Governance > Uniform Scrutiny Govern AI based on the actual operational risk created by what the system is allowed to do, rather than simply because it involves "AI". Low-risk administrative helpers shouldn't be bogged down by clinical-grade validation. Onboarding AI like a "New Employee" Traditional software quality assessments break when applied to open-ended, non-deterministic LLMs. Treating AI deployment more like onboarding, training, setting guardrails, and evaluating a new employee provides a practical framework for non-deterministic workflows. Solving the Client Data Paradox Clarifying model training boundaries vs. human context training, and establishing isolated data platform architectures to prevent IP leakage. Escaping "Pilot Purgatory" Building on a shared, governed platform foundation so life sciences leaders can scale AI from pilot to production without introducing compliance bloat. 🎧 Listen to the full episode now to rethink your AI validation playbook New Podcast @ctrlaltpod #EnterpriseAI #PlatformEngineering #GxP #LifeSciences #AIGovernance #Infiligence #PitchCafe #CtrlAltPodcast

#DF26 #hybridworkforce orchestration - CEO, Brandon Metcalf @Asymbl_Inc said its workforce runs 58% digital workers and 42% human employees across 13 functions. We dig deep on how he makes it work with @salesforce I.e. @SlackHQ #slackforce and #claudeforce @claudeai - #Salesforcepartner

How do banks run AI when communications data has to stay on campus? Rohit Khanna, Chief Customer Officer at Smarsh, on Ctrl Alt Podcast with @Vidyangi520 He says that constraint is already live. Bank communications are M&A, deals, contracts, and recorded CEO calls on Zoom or Teams. That is firm data. It does not go out to a public model. So the direction reverses. Models come to the bank. Many of those banks now run their own model factories. His product line for that is bring your own model: the institution keeps the compliance model, the workflow wraps around it. Same problem one level up, he says: country sovereignty, not only company sovereignty. Clip guest: Rohit Khanna, Chief Customer Officer, Smarsh Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: data sovereignty, bring your own model, bank communications archive, on-prem AI

Model risk management is the gap between an AI pilot and production. Question the clip answers: why do so many AI tools stall before a regulated firm will run them? Rohit Khanna, Chief Customer Officer at Smarsh, on Ctrl Alt Podcast with @Vidyangi520 He calls model risk management a big topic because experimentation to production is a different job. A lot of young AI companies fail that step, he says, whether they ship a frontier model or an open-source model. The model is not the production system. Production is whether the firm can govern the model. He still calls the opportunity the largest he has seen, and larger than anything he expects again in his lifetime. The constraint in this clip is the hop from lab to live. Clip guest: Rohit Khanna, Chief Customer Officer, Smarsh Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: model risk management, AI experimentation to production, frontier models, open source models

A bank with 100,000 employees could review 4 to 5% of its communications. AI is how they find the rest. Question the clip answers: why did misconduct surveillance break after the channel explosion, and what do regulators still require? Rohit Khanna, Chief Customer Officer at Smarsh, on Ctrl Alt Podcast with @Vidyangi520 He has not seen this velocity in three decades: time to market and time to value from AI. The case that was not possible before: • 100,000 employees at a large bank • Misconduct hunt by hand or by lexicons • Coverage stuck at 4 to 5% of the data • After Covid the channels multiplied: Zoom, Teams, Chatter, WhatsApp, WeChat, Signal • Volume up 100-fold • The needle in that stack, he says, is only findable with AI Regulated customers already know the value. The gap is responsible deployment. Experiment to production has hoops. You cannot drop a frontier model in front of FINRA or the SEC and treat the output as explained. Regulators still need to see how the model reached the conclusion. Clip guest: Rohit Khanna, Chief Customer Officer, Smarsh Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: AI misconduct surveillance, communications monitoring, FINRA, SEC, responsible AI

If you send an AI-written answer to a client, who is liable? Rohit Khanna, Chief Customer Officer at Smarsh, on Ctrl Alt Podcast with @Vidyangi520 . Regulators have already drawn the line, he says. Copy an AI answer into a client communication and the regulated person sent it. The engine is not liable. The agent is not liable. Where he thinks the market is going: • Regulated AI agents • If an agent gives wealth advice, that agent gets regulated • Its communications get recorded and surveilled • The test is accuracy, no misrepresentation, no misconduct He separates two speeds. AI company revenue has moved in a few years in a way he has not seen in 30 years in the industry. Governance still has to sit on top of that. His warning is governance that freezes the work. He flags talk of an FAA of AI and says the job is to keep regulators from pulling AI, and the economy, down. Clip guest: Rohit Khanna, Chief Customer Officer, Smarsh Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: AI liability, regulated AI agents, wealth advice surveillance, AI governance

When agents multiply inside a bank, do they need KYC? Rohit Khanna, Chief Customer Officer at Smarsh, on Ctrl Alt Podcast with @Vidyangi520. • More than 7000 AI experiments creating agents. Four reached production • Purpose-built agents first. Later, fewer super agents that can do more than one job • Enterprises cannot manage thousands of agent identities. He wants KYC on agents • Same controls used for people: single sign-on, multi-factor authentication, an Active Directory of named agents • 7000 to 8000 unmanaged agents is where he says leakage and breaches start Clip guest: Rohit Khanna, Chief Customer Officer, Smarsh Show: Ctrl Alt Podcast with Vida Patil Subjects discussed: AI agents, agent KYC, MFA, Active Directory, enterprise AI production

@CtrlAltpod Question the clip answers: do agents need KYC? Rohit Khanna: a top-five bank is running more than 7,000 AI experiments. Four are in production. The rest are identities with no owner. Purpose-built agents do not scale as a pile. They scale if each one is named, authenticated, and in a directory - @salesforce's Archie and @SmarshInc's Emmy, not a shared key. Full episode with @Vidyangi520 here piped.video/watch?v=9NxjyyqY…

Miro just sold for $1.355B. $17.5B in 2021. ~$600M ARR now. That’s ~2.3x revenue, 90% off the peak multiple. Question the clip answers: why did the 80% gross-margin SaaS machine stop printing those multiples? @priyasaiprasad, GP at @touringcapital , on @CtrlAltpod @Vidyangi520. Her argument: • Old-school SaaS worked because it was 80% gross margin, per-seat contracts, and operating leverage as you added users • AI puts inference and compute into COGS. She says you’re lucky at 40–50% • Per-seat pricing dies when the agent does the job. Fewer humans in the workflow = seat contraction, not expansion • Valuation math has to follow the P&L. Multiples, pricing, and “leverage” all have to adapt Miro is the exhibit. Collaboration software was priced as if every new teammate was a new seat forever. Bending Spoons is buying the cash-flowing version of that model after the market stopped paying 20x for it. Same clip. Same thesis. Today’s price tag.