The Efficiency Trap In AI

In this episode of TechUnhinged, host Ashar Samdani sits down with Chris Savage, CTO at ALIANDO, a Microsoft partner delivering cloud, data, and AI work in 30 countries. Chris argues that AI projects don't fail because of the technology; they fail when nobody ties them to a business process and a dollar figure. They cover AI implementation through process mapping, data permissions and hygiene, cloud versus on-premises AI infrastructure, and ROI metrics a CFO will accept. Chris also describes how an agent surfaced a spreadsheet of every employee’s salary that HR had shared without realizing anyone could open it.

In this episode of TechUnhinged, host Rabia Javeed sits down with Erik Gross, Cofounder at The Tech Academy, a former US Navy nuclear reactor operator and instructor. Erik argues that AI rarely fails on its own; it fails when leaders hand it work that needs identical outputs, uncleaned data, or untrained staff. They cover enterprise AI readiness, deterministic versus non-deterministic systems, AI literacy for non-technical staff, and what AI coding tools take away from new engineers. Erik also explains why 96 percent of employees whose unused SaaS accounts his team soft-deleted never asked for access back.

In this episode of TechUnhinged, host Rabia Javeed sits down with James Edgell, founder of SoftEXIT, a no-code application platform. James argues that AI is a strong coder but a poor planner, which makes clear requirements, data design, and architecture the work that now decides whether software holds up. They cover AI software development, feature drift in AI-generated code, the AI developer productivity gap, and the shrinking entry-level pipeline. James also explains why the DBA and the architect have become the most important people on a team once AI writes the code.

In this episode of TechUnhinged, host Ashar Samdani sits down with Andres Andreu, CEO of Constella Intelligence, an identity risk intelligence company. Andres argues that compliance doesn't make an enterprise secure, and that SOC 2 reports and written policies offer hope rather than protection once AI agents hold real privileges. They cover agentic AI security, AI agents as privileged identities, hybrid architecture for AI cost control, and deception engineering. Andres also describes how, by his account, Constella went from monitoring 400 Telegram channels to 11,000 using agents and without adding a single person.

In this episode of TechUnhinged, host Rabia Javeed sits down with Liza Adams, founder and AI advisor at GrowthPath Partners, a go-to-market veteran of five tech waves. Liza argues that AI adoption fails not because of the tools but because leaders skip the human change management that has to come first. They cover AI adoption, the mindset shifts that precede behavior change, risk matrix governance, and how AI search shapes buyer decisions. Liza also explains why the security team at PointClickCare became a co-builder rather than a blocker, and why that made every later workflow faster.

In this episode of TechUnhinged, host Ashar Samdani sits down with Joshua Nichols, Director of AI and Automation at CoorsTek, a technical ceramics manufacturer in Golden, Colorado. Joshua argues that access isn't adoption, and that people build confidence in AI through daily practice rather than license rollouts. They cover enterprise AI adoption, citizen development in Replit and Databricks, deterministic agents versus autonomous ones, and token budget caps. Joshua also explains why the biggest skill gap isn't prompting, but the willingness to reimagine a workflow around AI.

In this episode of TechUnhinged, host Rabia Javeed sits down with David (DJ) Morimanno, Field CTO for North America at Xalient, who advises Fortune 500 companies on identity. DJ argues that an AI agent should be governed like a digital worker, judged on why it acted, not only what it can access. They cover AI agent governance, privileged access management, agent ownership models, and the first moves for a company going AI first. DJ also explains why buying a tool before sizing the problem stalls identity programs, and why every user, human or not, now counts as privileged.

In this episode of TechUnhinged, host Ashar Samdani sits down with Sriram Ramanathan, Group CTO at XBP Global, a workflow automation company operating in 20 countries. Sriram contends that software companies selling licenses are already in trouble, and only those selling outcomes survive. He explains why the gap between agentic AI vs RPA is the difference between a flow that breaks when a button moves and a system that figures things out, why enterprises can skip the API rebuild and point agents at the mess they already have, and how his team now measures agent productivity the same way it measures a human's. He also describes the cluster of nine agents that rewrote his own development process.

In this episode of TechUnhinged, host Rabia Javeed sits down with Risto Uuk, Head of European Policy and Research at the Future of Life Institute, who has spent five years on the Act's general-purpose AI rules. Risto argues that the rules for the biggest models are not the weak point; the open question is whether anyone will enforce them. He explains why training compute is a proxy for danger rather than a measure, why the Code of Practice was built from the labs' own safety frameworks, and how the rules reach a company with no office in Europe. He also walks through the agent that broke out of a test environment and hacked third-party infrastructure.

In this episode of TechUnhinged, host Rabia Javeed sits down with Charath Ranganathan, Chief Technology Officer at CoreStory, which builds tools that pull specifications out of legacy codebases. Charath argues that AI has replaced coding and has not affected the discipline of engineering, and that conflating the two is the most misleading assumption enterprises bring into an AI rollout. He explains why a 1970s mainframe resists modernization because code records what was decided, not why; why evals belong in front of production rather than behind it; and how a flood of generated code produces cognitive surrender. He also goes back to the pixelated circle on a Sinclair Spectrum that made him learn trigonometry on his own.