Inside AsembleAI: DeepTech, AI & Science

AI moves fast; laws struggle to keep up. In this episode of Inside Assemble AI, Mac Goswami and Sam Dey tackle the most pressing questions about the future of AI policy—from Artificial General Intelligence (AGI) that could exceed human capabilities to the murky liability questions around autonomous AI agents.

What happens when AI agents cause harm? Who's liable - the developer, the deployer, or the user? Current regulations weren't designed for systems that can make independent decisions, negotiate contracts, or interact with other AI systems. The legal framework is unclear and complex, and we're already behind.

The episode explores the double-edged sword of open source AI: it fosters innovation and democratizes access, but it also complicates control and regulation. How do you govern models that anyone can download, modify, and deploy? The traditional regulatory playbook doesn't work when the technology is freely distributed.

Key insight: "AI policy will evolve as rapidly as AI itself." This isn't a one-time regulatory fix—it's a continuous process of adaptation, learning, and cooperation. Current regulations are already inadequate for AGI scenarios, and we need frameworks that can flex with technological advancement rather than break under it.

The conversation emphasizes that public participation is crucial in shaping AI policy. These decisions affect everyone, and the dialogue can't be left only to technologists and policymakers.

Topics covered: AGI implications for humanity, AI agent liability frameworks, open source AI governance paradox, synthetic content detection and regulation, global cooperation mechanisms, technology governance evolution, continuous regulatory adaptation

Subscribe to Inside Assemble AI where AI, deep tech, and science meet storytelling. Stay curious and build responsibly.

The EU has its AI Act. The US has Biden's executive order followed by AI Action Plan released last year. China has something entirely different. In this episode, Sam and Mac zoom out to examine the global landscape of AI regulation—and it's not just about different rules, it's about competing visions of technology and society.

What you'll learn:

• US sectoral approach: Different agencies (FDA, FTC, EEOC) regulate AI in their domains—flexibility but fragmentation
• China's radically different model: Algorithm registration, content filtering aligned with socialist values, state oversight
• Middle-path approaches: UK's pro-innovation framework, Canada's EU-aligned AIDA proposal, Singapore's voluntary incentives
• Is the Global South being left behind? Risk of regulatory colonialism from Brussels and Washington
• Regulatory convergence vs fragmentation: Shared principles (transparency, accountability, fairness) but wildly different implementation
• Data localization challenges: China, Russia, Indonesia require local storage—making global AI models harder to train

Critical flashpoints:

• Content moderation: What counts as "harmful" varies drastically by country
• Technical standards: ISO, IEEE, NIST developing frameworks, but who sets standards matters geopolitically
• Market fragmentation: Chinese AI companies don't operate in the West; Western companies avoid China

For AI builders and startups: Design for the most stringent requirements you expect. Build in privacy, transparency, and accountability from the start. If you want EU customers, you comply with EU rules—regardless of where you're based. Focus on your target market first for validation, then expand compliance as you scale.

Key insight: These aren't just regulatory differences—they're geopolitical choices that shape what gets built, how it works, who benefits, and what risks we accept.

Data governance isn't sexy, but it's what makes or breaks your AI strategy. In this episode, Sam and Mac tackle the tactical reality of what happens inside companies trying to comply with AI regulations while keeping data governance practices intact.

What you'll learn:

• Why you can't have compliant AI without proper data governance
• Data lineage: tracking where your data came from, how it's processed, and where it ends up
• Real-world bias example: How historical hiring data can violate EU AI Act principles
• The challenge of GDPR's "right to be forgotten" when data is baked into neural networks
• Model governance across the entire lifecycle—from selection to deployment monitoring
• Why human oversight remains critical in high-risk systems like loan decisions
• How smaller companies can stay compliant without enterprise-level budgets

Key frameworks covered:

✓ Data lineage and chain of custody

✓ Audit trails throughout the AI lifecycle

✓ Model cards for documentation (used by Google, Microsoft, Meta, Amazon)

✓ Post-deployment monitoring: data drift, concept drift, and bias detection

✓ Human-in-the-loop requirements for consequential decisions

The unsexy truth: Compliance as a service companies are emerging to help startups navigate these requirements. Trust isn't just a nice-to-have—it's becoming a competitive advantage.

The EU AI Act became law in 2024, and even if you're not in Europe, it's going to affect how you build with AI. In this episode, Sam and Mac break down the world's first comprehensive AI regulation—from banned applications to high-risk use cases that require strict oversight.

What you'll learn:

• The four-tier risk framework: unacceptable, high, limited, and minimal risk
• Why this matters for your AI projects (hint: think GDPR's global impact)
• How enterprises balance innovation with compliance
• Practical implementation strategies from the frontlines
• What "the right to be forgotten" means when data is baked into neural networks

Whether you're building AI applications, leading data teams, or navigating enterprise AI governance, this episode gives you the framework to implement AI responsibly while maintaining innovation velocity.

Timeline rollout: Bans effective early 2025, general purpose AI requirements mid-2025, full high-risk compliance by mid-2026.

Mac and Sam break down Model Context Protocol (MCP)—the universal standard transforming how AI connects to tools, data, and workflows. Think of it as the "USB-C moment" for AI: plug-and-play integration that eliminates custom builds for every system.

Discover what MCP actually is, how it enables seamless AI connections across your tech stack, real-world use cases for developers and enterprises, and why regulated industries are taking notice.

Key Topics:

• The shift from custom integrations to standardized protocols
• Practical implementation strategies
• Enterprise and regulated industry applications
• What MCP solves (and its limitations)

Perfect for: AI developers, enterprise teams, tech leaders in regulated industries, and anyone curious about the future of AI tooling.

Key Insight: MCP isn't about making AI smarter—it's about making AI connections smarter.

Resources: MCP documentation, weekly insights, and community links at asembleai.substack.com

Discover why the future of AI isn't one "super agent" but coordinated teams of specialized agents working together. We explore Agent-to-Agent (A2A) communication—the protocol enabling AI agents to collaborate on complex tasks.

Learn how A2A frameworks are transforming AI from isolated demos into production-ready organizations, the essential components of multi-agent systems, and why oversight mechanisms are critical for deployment.

Key Topics:

• Why single agents fail at complex workflows
• A2A protocols and collaboration frameworks
• Building production-ready agent teams
• Real-world applications and risk management

Perfect for: AI engineers, technical leaders, product managers, and anyone building with AI agents.

Resources: A2A Framework documentation, weekly AI insights newsletter, and community links at asembleai.substack.com

2025 wasn't just about smarter models—it was about an entire ecosystem evolving around us. Developer toolkits matured, Big Tech made billion-dollar bets, and AI went from experiment to production infrastructure across every major industry.

In this episode, Mac Goswami and Sam Dey break down what actually happened: the frameworks that made AI development accessible, Google vs. Microsoft vs. Meta's strategic plays, and how healthcare, finance, education, and creative industries are being transformed—with appropriate guardrails.

Topics include the maturation of AI dev platforms like Vercel AI SDK, LangChain, and LlamaIndex that compress weeks of work into hours; vector databases and RAG frameworks hitting production readiness; why observability tools became the most underrated development of 2025; Google's aggressive Gemini push, Microsoft's Copilot-as-platform strategy, and Meta's open-source gambit with Llama; Amazon's infrastructure play and the GPU vs. TPU battle; healthcare AI for diagnostics, drug discovery, and clinical documentation with human oversight; finance applications including fraud detection, portfolio analysis, and compliance reporting; the EdTech dilemma of AI literacy vs. academic integrity; startup survival strategies when Big Tech enters every space; and 2026 predictions covering multimodal integration, on-device AI, and continued consolidation.

Key insight: AI moved from experimentation to production at scale. The augmentation pattern won over the replacement narrative—across every industry, successful AI applications enhanced human expertise rather than replacing it.

Hosts: Mac Goswami and Sam Dey

Show: AsembleAI - Where AI, Deep Tech & Science meet storytelling

Connect: https://substack.com/@asembleai? | https://www.linkedin.com/company/asembleai

The AI landscape shifted dramatically in 2025. Agents aren't just executing commands anymore—they're making decisions, coordinating tasks, and working alongside humans as true collaborators.

In this episode, Mac Goswami and Sam Dey break down what actually changed: expanded context windows, reasoning models that finally work, standardized tool use, and multi-agent frameworks ready for production. They also dig into what went wrong, the failures that taught the industry hard lessons about engineering and oversight.

Topics covered include the technical breakthroughs that made 2025 the inflection point for agent AI, real-world applications proving agents are production-ready, how to calibrate autonomy to match the stakes, single vs. multi-agent systems, framework maturity, cautionary tales from AI failures, and trends shaping 2026.

Key insight: Start simple. Decompose when needed. The future isn't AI replacing humans—it's human-agent collaboration done right.

Hosts: Mac Goswami and Sam Dey

Connect: https://www.linkedin.com/company/asembleai

In this episode of Inside Assemble AI, hosts Mac and Sam sit down with Priya Reddy, an expert in AI strategy and data literacy, to explore why the real bottleneck in AI transformation isn't technology, it's understanding!

Priya shares her insights on the critical importance of AI literacy across organizations, emphasizing that successful AI adoption requires bridging the gap between technical teams and business leaders. The conversation delves into why organizations must invest in people rather than just tools, how translation failures between data scientists and executives derail AI initiatives, and what it takes to build a culture of data-driven decision-making.

From real-world examples of AI adoption failures to practical frameworks for effective data literacy programs, this episode offers actionable guidance for leaders navigating the rapidly evolving AI landscape.

Key Topics Discussed:

• Why AI literacy—encompassing data, tech, and business literacy—is the foundation of successful AI transformation
• Real-world examples of how miscommunication between technical and business teams derails AI initiatives
• The difference between data quality issues and AI model problems, and why it matters
• How executives can ask the right questions about data and AI to drive better outcomes
• Building role-based, practical data literacy programs that stick
• The role of "bilinguals"—professionals who can translate between technical and business contexts
• Why data-literate leaders will define the next era of competitive advantage

About Our Guest:

Connect with Priya on LinkedIn: https://www.linkedin.com/in/priyareddy-dataleader/

Checkout her consultancy services: https://www.linkedin.com/company/priya-reddy-consulting/

In this episode of Inside Assemble AI, hosts Sam Dey and Mac Goswami speak with Sam Bhagwat, Co-Founder & CEO of Maastra, about the rise of AI agents and the tools needed to build them for production.

Sam shares his journey from Gatsby to Maastra, the developer pain points that inspired the pivot, and why context engineering is critical for effective AI agents. The conversation explores Maastra’s core primitives, the impact of the Model Context Protocol (MCP), and the challenges of moving from prototype to production. The episode concludes with a forward-looking vision of AI development—smaller models, broader access, and a future with a billion developers by 2030.

🔗 Guest Links

• LinkedIn: https://www.linkedin.com/in/sambhagwat/

• Mastra: https://mastra.ai/

• Book(s): https://www.amazon.com/Principles-Building-Agents-Sam-Bhagwat/dp/B0DYH5GHDD