Mythos, Legacy Contracts and the Fog Premium | E83

AI Is Changing Engineering Jobs Fast | Brian Samson

AI is changing software engineering faster than most teams are prepared for. AI coding tools are generating code faster than engineers can review it. Junior hiring is slowing down. And the value of engineering work is shifting from pure execution toward validation, judgment, ownership, and context. In Episode 75 of The Spark & The Forge, Subrata Kar speaks with Brian Samson — founder of Plugg Technologies and recruiter placing engineering talent across Latin America and US tech companies. This conversation explores: why CTOs are still hiring despite AI coding tools * the hidden “rework tax” behind AI-generated code * why demand is shifting toward senior engineers * deepfakes and bots in technical interviews * India’s engineering transition * the rise of the “full stack unit” TIMESTAMPS This is not a doom conversation about AI replacing engineers. It is a conversation about what becomes more valuable as AI improves. 00:00 — “I think that company is not gonna make it” 00:16 — The shift nobody is naming clearly yet 01:08 — Why CTOs are still hiring despite AI coding tools 01:54 — The hidden rework tax of AI-generated code 02:44 — Why demand is shifting toward senior engineers 03:37 — Bots and ghost candidates in interviews 06:33 — What this means for India’s engineering pipeline 08:38 — Why YC startups are hiring engineers in Latin America 12:19 — “Become the full stack unit” 13:27 — The window is still open. But not indefinitely. Brian Samson https://www.linkedin.com/in/briansamson/ Founder — Plugg Technologies Host — The Nearshore Cafe Podcast Subrata Kar https://www.linkedin.com/in/subroto/ Host — The Spark & The Forge GUESTHOST

Mythos, Legacy Contracts and the Fog Premium | E83

In April 2026, Anthropic released Mythos — an AI model that read a 30-year-old system and found what humans missed for 27 years. Autonomously. This is not a security story. It is a contracts story. Every managed-services SLA running right now was written for a world where human diligence was the best available standard. Mythos just changed what best available means overnight. In this episode I share what I see in this — and how I am thinking about what it means for legacy IT contracts, Indian IT services, and the standard of care embedded in managed services agreements. What you will hear: — The three things Mythos demonstrably did: OpenBSD, FFmpeg, binary reconstruction — Why this is not a security story — it is a contracts story — The three opportunities Mythos just made viable that did not exist 60 days ago — The legal standard of care shift and what Gilbert + Tobin said about board liability — Why Berkshire Hathaway, Chubb, and Travelers excluded AI damages from general liability — The fog premium question — how much of your margin was genuine expertise? — What AI deflation means and why HCL and TCS are already naming it — The Trust Premium framework — where value moves when production gets cheap Host: Subrata Kar The Spark & The Forge — patterns from practitioners. Newsletter: https://substack.com/@subratakar YouTube: https://youtube.com/@thesparkandtheforge LinkedIn: https://www.linkedin.com/in/subroto SOURCES Anthropic Project Glasswing: https://www.anthropic.com/glasswing Anthropic Red Team Blog: https://red.anthropic.com/2026/mythos-preview/ Gilbert + Tobin Legal Analysis: https://www.gtlaw.com.au/insights/how-mythos-class-ai-is-changing-the-cyber-security-risk Mozilla Firefox 271 Vulnerabilities: https://thenextweb.com/news/mozilla-firefox-claude-mythos-271-vulnerabilities Insurance Exclusions: https://www.pymnts.com/artificial-intelligence-2/2026/big-insurance-backs-away-from-ai-risk APRA CPS 234: https://www.apra.gov.au/information-security NASSCOM Letter: https://www.medianama.com/2026/04/223-india-anthropic-claude-mythos-project-glasswing-access/ OpenAI Daybreak: https://www.techbuzz.ai/articles/openai-just-released-its-answer-to-claude-mythos AI Deflation: https://www.theregister.com/2026/04/28/tcs_infosys_wipro_hcl_fy26/ Coalition CEO Quote: https://www.coalitioninc.com/blog/cyber-insurance/after-mythos-what-actually-changes-for-cyber-risk DISCLAIMER This podcast is for informational and educational purposes only. It does not constitute professional financial, investment, or legal advice. Always conduct your own research and consult with qualified professionals before making business, legal, or financial decisions. FAIR USE NOTICE Referenced materials in this episode are used for purposes of commentary, analysis, criticism, and education under Section 107 of the Copyright Act (Fair Use).

$15M SaaS Vendor GONE in Months | What Chamath Built | E81

Software is getting 60-80% cheaper to build. That's changing everything about buying versus building enterprise software. Last September, Chamath Palihapitiya made a bold claim: one company is replacing a $15 million-per-year SaaS vendor using what he calls a Software Factory. Not switching to a competitor. Replacing it entirely. Here's what's happening beneath the hype: SaaS valuations crashed from 20x revenue to 3.5x. India's IT services revenue grew 6.1% while headcount grew only 2.3%—the widest gap in a decade. Big Four consulting firms are racing to deploy Software Factory services. When production gets industrialized, economics shift. The question is: where does the value move? In this episode, I break down: - Why Software Factories work NOW (the GenAI breakthrough that changed everything) - How AI maps 30-year-old legacy systems we thought were unmappable - The 4-step factory process (map → knowledge graph → assembly → validation) - Why India IT firms face pricing pressure first (offshore model disruption) - Why engineering jobs grow 17% while routine coding declines (BLS data) - The Trust Premium: why validators win, not the fastest builders This isn't about coding faster. It's about reducing the cost of understanding, changing, and validating enterprise software. That's the real breakthrough. TIMESTAMPS: 0:00 Hook: Software Getting 60-80% Cheaper 0:30 Chamath's $15M SaaS Replacement Claim 1:15 Three Signals: SaaS Crash, India IT Gap, Big Four 2:30 Category Formation (EY Partnership) 3:25 Why DevOps, Offshore, Low-Code Didn't Change Economics 5:15 India IT: NASSCOM 6.1% vs 2.3% Data 6:25 Why FY26 Is Different (AI Pilots → Production) 8:20 SaaS Valuation Collapse (20x → 3.5x) 9:45 Legacy Fog: 80% Investigated AI, 5% in Production 11:30 Why Traditional Approaches Fail 12:45 Software Factory 4-Step Process 14:15 Why Couldn't We Do This Before? (GenAI Breakthrough) 15:50 JPMorgan COiN Example (360K Hours Saved) 16:45 BCG 70% Rule: People, Process, Change Management 18:05 Engineering Jobs: Growing 17%, Not Declining 18:50 The Rebuttal: Global Growth vs India Decline 19:50 Trust Premium: Validators Win, Not Builders --- FULL ANALYSIS: Read the detailed breakdown on my Substack covering who loses margin first, three pricing models emerging, and the complete reconciliation of why global jobs grow while India headcount declines. Link: https://substack.com/@subratakar --- Host: Subrata Kar The Spark & The Forge I study patterns from builders who scale—enterprise systems, AI platforms, and startups—and extract actionable insights leaders can apply immediately. LinkedIn: https://www.linkedin.com/in/subroto Newsletter: https://substack.com/@subratakar YouTube: https://youtu.be/8pWI3W-M4xM

The Trillion-Agent Problem Nobody's Talking About

Insurers have stopped covering AI failures. Claudionor Coelho Jr. — ex-Chief AI Officer at Zscaler & Avantest, Stanford PhD — breaks down why. 95% of AI pilots never make it to production. Not because the AI fails. Because the security breaks. The reliability breaks. The governance breaks. Something else breaks first. Claudionor has spent 30 years building mission-critical systems — from chip-level verification at Stanford to 500 billion daily transactions at Zscaler to CERN's Large Hadron Collider. He's now Senior AI Fellow at Majestic Labs, advising Daxa on reasoning layers and Ridge Security on agentic pentesting. In this conversation we cover: — Why AI risk has become revenue risk— The Swiss Cheese problem with traditional security in agentic systems— Why zero trust protects the connection but not the behavior— The Autonomy Paradox — why "just have humans check the output" fails— True Zero Trust explained — from semiconductor manufacturing to enterprise AI— The math of multi-agent reliability: 90% accurate agents produce a system that works 35% of the time— Why triple redundancy fails when errors are correlated (the Air France lesson)— The auditability gap in multi-agent systems— Reasoning layers (Daxa) and agentic pentesting (Ridge Security)— The trillion-agent future and Project NANDA at MIT— Why AI-powered social engineering is now nearly impossible to detect 00:00 — Introduction02:48 — Building agents is a software engineering problem, not an LLM problem05:02 — APIs, trillions of agents, MCP, and flood gates to hell08:17 — The Swiss Cheese problem11:34 — Probabilistic vs discrete systems — why LLMs hit hard limits14:50 — Origin of True Zero Trust18:33 — The Autonomy Paradox — Tesla & MIT studies21:35 — Vibe coding danger — the lab deleted by "Yes to All"23:26 — Data leakage reality — the red and blue customer26:34 — Business rules that must be 100% guaranteed28:14 — Multi-agent failure math: 90% → 35% + Air France32:03 — The auditability gap35:16 — Reasoning layers — Daxa / Pebblo neuro-symbolic approach38:51 — Ridge agentic pentesting — attacks humans can't conceive42:23 — Project NANDA — the internet of agents47:01 — Personal attacks using world understanding47:11 — Key takeaways "I call this flood gates to hell. You open the gate and you start sending confidential information to that MCP connection. And that MCP connection is people on the other side." — Claudionor Coelho Jr. GUESTClaudionor Coelho Jr.Senior AI Fellow — Majestic Labsex-Chief AI Officer at Zscaler & Avantest | ex-Google | Stanford PhDAdvisor to Daxa & Ridge SecurityLinkedIn: https://www.linkedin.com/in/claudionor-coelho-jr-b156b01 [https://www.linkedin.com/in/claudionor-coelho-jr-b156b01] HOSTSubrata Kar — The Spark & The ForgeI study patterns from builders who scale — enterprise systems, AI platforms, and startups — and extract actionable insights leaders can apply immediately.LinkedIn: https://www.linkedin.com/in/subroto [https://www.linkedin.com/in/subroto]Newsletter: https://substack.com/@subratakar [https://substack.com/@subratakar] Full episode also on YouTube: https://youtu.be/eE7lYEtXHIQ [https://youtu.be/eE7lYEtXHIQ]

AI Drift — How Companies Lose Control Without Deciding|E72

Most companies didn't choose AI dependency. They drifted into it — one tool at a time, one workflow at a time.In this episode, I speak with Tamsin Deasey-Weinstein, who leads AI strategy for the Cayman Islands — a financial hub thinking seriously about AI sovereignty, governance, and capability.We explore:- Why AI adoption fails even in well-funded organizations- Why the real bottleneck is talent, not compute- The near-zero-cost training model that scales- Why governance enables speed (not slows it)- How companies lose leverage when workflows depend on "rented intelligence"- The real risk of AI: not job loss, but loss of judgmentThis isn't a tools episode. It's about drift, dependency, and deciding before it's too late.🔗 CONNECTGuest: Tamsin Deasey-WeinsteinLinkedIn: https://www.linkedin.com/in/tamsindeaseyweinstein/Host: Subrata KarNewsletter: https://www.linkedin.com/newsletters/7240332159192875008LinkedIn: https://www.linkedin.com/in/subroto/The Spark & The Forge — Patterns That Actually WorkExtracting scaling patterns from leaders building enterprise, AI, and future-of-work systems.