Podcast
The CTO Show with Mehmet Gonullu
Hosted by Mehmet Gonullu · EN
Broadcasting from Dubai, The CTO Show with Mehmet explores the latest trends in technology, startups, and venture funding. Host Mehmet Gonullu leads insightful discussions with thought leaders, innovators, and entrepreneurs from diverse industries. From emerging technologies to startup investment strategies, the show provides a balanced view on navigating the evolving landscape of business and tech, helping listeners understand their profound impact on our world.
mehmet@yassiventures.com
51episodes
Episodes
Newest firstAll episodes
#609 AI Can Assess Leaders. It Shouldn’t Replace Judgment | Logan Yonavjak
3d ago00:51:31Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Logan Yonavjak, Co-Founder and CEO of Founder Readiness Engine. Logan brings an investor and operator view into how founders and senior leaders can be assessed beyond resumes, charisma, and gut feel.The conversation reframes leadership assessment as a decision system, not a personality test. Logan explains how transcript data, developmental psychology, quantitative linguistics, and AI can surface signals such as coachability, identity flexibility, strategic complexity, relational intelligence, and resilience. The key tension is clear: AI can improve how leaders are assessed, but humans should not hand over agency to the machine.If you are investing in founders, hiring senior leaders, building leadership teams, or evaluating startup risk, this conversation gives you a sharper way to think about people analytics, founder readiness, and AI-assisted decision-making.About the GuestLogan Yonavjak is the Co-Founder and CEO of Founder Readiness Engine. She is an impact investor turned entrepreneur with experience across private equity, university endowments, farmland investing platforms, sustainable investing, and early-stage technology.She teamed up with a data scientist and psychologist to build a platform that analyzes transcript data and identifies leadership readiness markers. Her work focuses on how founders, senior leaders, investors, and organizations can make better decisions about people under pressure and complexity.LinkedIn: https://www.linkedin.com/in/loganyonavjak/Website: https://www.readinessengine.io/Key TakeawaysAI can assess leadership readiness, but it should not replace human judgment.Founder evaluation still depends too heavily on gut feel, charisma, and warm references.Coachability and identity flexibility are critical signals for founder growth.Traditional assessments often miss how leaders develop under pressure and complexity.Strategic complexity shows up in how leaders hold multiple perspectives at once.Resilience is not a trait alone, it is a system leaders build around themselves.Relational intelligence can offset blind spots in highly technical or visionary founders.People analytics may become a stronger diligence layer for investors and operators.What You Will LearnHow AI can analyze transcript data to identify leadership readiness signals.Why coachability may matter more than credentials in founder evaluation.The limits of traditional assessments such as MBTI, DiSC, StrengthsFinder, and Predictive Index.How strategic complexity appears in the way leaders explain systems and tradeoffs.Why human agency must remain central when AI supports hiring or promotion decisions.What investors often miss when they rely on pattern matching and warm references.How leadership assessment could become part of due diligence, hiring, and lending decisions.Episode Highlights00:00 — Founder readiness becomes the central question05:00 — Traditional assessments miss developmental trajectory09:00 — Negative space reveals what leaders avoid13:00 — AI should augment, not replace judgment20:00 — Coachability becomes the strongest founder signal23:00 — Relational intelligence offsets leadership blind spots29:00 — Strategic complexity appears in language patterns31:00 — Resilience depends on systems under stress35:00 — Leadership data could reshape lending decisions39:00 — VC still relies heavily on gut checks43:00 — AI can model a stronger second brain46:00 — Technology can uncover human blind spotsListen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#608 AI Won’t Replace Therapists. It May Replace Guesswork | Dr. Steve Rondeau
6d ago00:54:54Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Dr. Steve Rondeau of AxonEG Solutions. Dr. Steve brings more than two decades of work across developmental medicine, EEG brain scans, biomarkers, and mental health diagnostics. The core tension is clear: mental health has too often treated labels as answers, while the brain may be telling a different story.The conversation reframes AI in healthcare as a decision-support layer, not a replacement for clinicians. Dr. Steve explains why two people with the same diagnosis can respond completely differently to treatment, how a database of more than 50,000 brain scans changes the conversation, and why objective biological data can reduce trial and error in care. The episode also connects AI, explainability, human judgment, and empathy in a field where the cost of guessing can be very high.If you are building, investing in, or leading in AI, healthcare technology, digital health, or human performance, this conversation shows where data can improve decisions without removing the human from the loop.About the GuestDr. Steve Rondeau is with AxonEG Solutions, where his work focuses on EEG brain scans, biological markers, and objective data in mental health diagnostics. He is also the author of Think Like a Brain, a book focused on helping people understand brain patterns, treatment response, and why labels alone do not explain the full picture.His work is built around a database of more than 50,000 brain scans and a central question: why two people with the same mental health diagnosis can respond so differently to treatment.LinkedIn: https://www.linkedin.com/in/dr-steven-rondeau-148aa421/Website: https://thinklikeabrain.comKey TakeawaysMental health labels describe suffering, but they often fail to predict treatment outcomes.AI can support clinicians by narrowing options, not by replacing human judgment.A single diagnosis can hide thousands of possible biological patterns.Objective brain data can reveal treatment paths that symptom labels may miss.The DSM helps clinicians communicate, but it does not explain each patient’s biology.Human-in-the-loop AI matters most when decisions involve context, culture, and empathy.Personalized mental health requires testing the organ being treated.Psychedelic and neuromodulation treatments need better prediction before wider adoption.What You Will LearnThe reason symptom-based diagnosis can miss the biological drivers behind treatment response.How EEG brain scans can add objective data to mental health decisions.Why two patients with the same diagnosis may need completely different treatments.The role AI can play in connecting biomarkers, clinical data, and published research.How human judgment remains essential when algorithms recommend clinical paths.Why treatment prediction matters for psychedelics, ketamine, and neuromodulation.What personalized medicine looks like when the brain is measured directly.Episode Highlights00:00 — Why mental health needs better data02:30 — Diagnosis describes symptoms, not treatment outcomes07:30 — Building a 50,000 brain scan database12:30 — One diagnosis can hide thousands of patterns16:30 — A brain scan challenged the symptom label20:00 — Brain data can open harder conversations23:30 — Biology and environment shape the same brain28:30 — AI supports clinicians, not replaces them31:30 — Predicting who responds before treatment starts35:00 — Psychedelics need better patient selection40:00 — Mental health should test the organ it treats45:30 — Adoption depends on validation, funding, and trust52:30 — Where to find Dr. Steve RondeauListen NowAvailable on all major podcast platforms and YouTubeConnect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#607 More AI Won’t Help. Better Processes Will | Lara Hamilton
1w ago00:37:23Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Lara Hamilton, a technology leader at HelpDesk Realty. The conversation focuses on why more AI will not help companies that have not fixed their processes first.Lara reframes AI adoption as an operations problem rather than a technology problem. The discussion moves from property management and paperless workflows to AI agents, security, documentation, and the practical friction that slows teams down. The core argument is clear: AI can save time, but only when the business knows how the work actually gets done.If you are leading IT, operating a growing business, investing in enterprise technology, or evaluating AI projects, this conversation gives a grounded view of where automation works and where it breaks.About the GuestLara Hamilton is a technology leader at HelpDesk Realty, where she works across IT operations, support, property technology, and compliance.Her background includes banking operations, process improvement, help desk services, property management systems, cybersecurity practices, and practical AI adoption.Lara brings an operator’s view of AI because she works with the systems, users, reports, tickets, and workflows that determine whether technology succeeds or fails.LinkedIn: https://www.linkedin.com/in/larahamilton-multifamilyit/Website: https://www.teamtectonic.com/divisions/helpdesk-realtyKey TakeawaysAI does not fix broken processes, it depends on them being clear first.Undocumented work becomes a major risk when companies try to automate it.Operational friction is often where AI produces the clearest return.Small daily tasks can create large productivity losses when repeated across teams.AI agents can block support when they replace human escalation paths.Security controls fail when users experience them as constant friction.Multi-factor authentication remains unpopular, but it is still necessary.Human knowledge inside teams cannot be replaced by tools alone.What You Will LearnHow missing process documentation weakens AI adoption.Why AI projects fail when leaders start with tools instead of workflows.The specific types of operational friction that automation can remove.How ticketing data and reporting tasks can become practical AI use cases.Why AI agents still need human escalation paths.When security controls improve protection without hurting productivity.What property management can teach broader enterprise teams about digital adoption.Episode Highlights00:00: Lara Hamilton’s path from banking operations to IT02:00: Repetition creates the strongest case for automation04:00: Property management still carries manual process debt05:30: Paperless workflows expose resistance to operational change08:30: IT leaders must translate vision into execution09:30: Undocumented processes block better technology outcomes12:00: AI depends on the foundation beneath it16:30: Small AI use cases can return hours weekly22:00: AI agents can break support escalation paths25:30: Security must balance protection with user behavior28:00: Digital payments changed property operations after COVID31:00: Strong IT teams share knowledge across skill sets33:30: Learning compounds into institutional knowledgeListen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#606 AI Can Generate Code. It Still Can’t Replace Engineering Judgment | Jason Li
2w ago00:45:13Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Jason Li, CTO at Laurel. Jason brings experience from enterprise software, Salesforce, Ironclad, and AI-native product development.The conversation reframes AI adoption away from replacing work and toward understanding work. Faster code generation does not eliminate engineering bottlenecks. Quality, technical debt, review processes, and organizational design are becoming the limiting factors.If you are leading engineering teams, building AI products, or investing in enterprise software, this conversation provides a practical view of how AI is changing software development and technical leadership.About the GuestJason Li is the CTO at Laurel, an AI company focused on time intelligence and productivity. Previously, he worked in enterprise software and held roles at Salesforce and Ironclad.His work spans AI-native products, developer productivity, legal technology, and engineering leadership.His perspective comes from operating AI systems inside production environments while managing the realities of software quality, technical debt, and team structure.LinkedIn: https://www.linkedin.com/in/jasonhli/Laurel website: https://www.laurel.ai/Key TakeawaysAI shifts bottlenecks from code generation to code quality.Visibility into work creates more leverage than blindly automating tasks.Engineering productivity remains difficult to measure despite new AI tools.Agentic coding increases the speed at which technical debt accumulates.Existing code review processes were not designed for AI-generated code.Senior engineering judgment becomes more valuable in an agent-driven world.AI tools expose weaknesses in processes rather than eliminating them.Rewriting software may become cheaper and more common than in previous generations.What You Will LearnThe difference between replacing work and understanding work.How time intelligence creates operational visibility.Why measuring AI ROI remains difficult.How engineering teams are adapting to agentic coding.What skills remain valuable for engineers entering the profession.Why technical debt may increase faster in AI-assisted development.When software rewrites may become preferable to maintaining legacy architectures.Episode Highlights00:00 — Time intelligence extends beyond billing hours03:30 — Visibility matters before automation decisions05:00 — AI should amplify leverage, not replace people08:00 — Trust and reliability determine AI adoption12:00 — AI systems inherit organizational weaknesses15:00 — Measuring AI productivity remains difficult17:30 — Agentic coding changes software engineering20:00 — Engineering leadership becomes more hands-on25:00 — Judgment matters more than coding syntax30:00 — Technical debt grows faster with AI35:00 — Wrappers versus foundation model tools40:30 — Uncertainty creates new opportunitiesListen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, AI infrastructure, cybersecurity, and venture capital.
Transcribe →#605 AI Won’t Fix Broken Organizations. It Exposes Them | Jürgen Dauk
2w ago00:53:38Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Jürgen Dauk, advisor, consultant, and creator of the Leadership Operating System. AI is not the real bottleneck. Broken organizational design is.The conversation reframes AI adoption as a leadership and operating model problem rather than a software rollout. Jürgen argues that companies built around control, reporting, and top-down approval are too slow to capture real value from AI. The discussion moves from misaligned KPIs and forecast calls to distributed decision-making, experimentation, and why AI often amplifies the dysfunction already inside the company.If you are leading, investing in, or operating an enterprise technology company, this conversation clarifies why AI value depends less on tools and more on how decisions, teams, and accountability are designed.About the GuestJürgen Dauk is an advisor and consultant to companies and the creator of the Leadership Operating System. He is the author of The Leadership Operating System and has worked across technology, marketing, sales, customer support, customer success, and management roles.Jürgen’s background includes work with companies such as Oracle and OpenText, as well as transformation work across mid-sized and large organizations. His work focuses on helping companies move away from fear-based control and toward operating models where people, teams, and decision-making can support faster adaptation.LinkedIn: https://www.linkedin.com/in/juergendauk/Website: https://theleadership-os.com/Key TakeawaysAI does not fix broken organizations. It makes their weak points more visible.Company-wide AI rollouts fail when leaders mistake access for adoption.Control-based operating models create stability, but they also slow decision-making.Misaligned KPIs push sales, marketing, and customer success into internal conflict.AI should not automate bad processes before leaders question why those processes exist.Distributed decision-making becomes a survival issue when competitors move faster.Reporting calls and alignment meetings often create activity without real output.AI can multiply low-value work when organizations use it to produce more noise.What You Will LearnThe organizational patterns that prevent companies from benefiting from AI.Why Microsoft Copilot access alone does not create measurable productivity gains.How leaders can move from centralized AI rollouts to team-level problem solving.The role of distributed decision-making in faster AI adoption.Why experimentation culture matters more than formal AI training.How reporting calls, CRM inspection, and dashboards can create false control.What leadership teams must change before AI can create real operational value.Episode Highlights00:00 — AI exposes the organization behind the tooling05:00 — Misaligned KPIs turn teams against each other09:00 — Command and control was built for stability15:00 — Company-wide AI rollout can produce little value17:00 — AI works when teams rethink the process20:00 — Technical expertise belongs inside business teams22:00 — Experimentation turns failed pilots into useful learning25:00 — Reporting calls create alignment without real output29:30 — AI can multiply nonsense work38:30 — Slow decisions are now existential risk43:30 — The Leadership Operating System connects the pieces51:00 — Jürgen shares resources for organizational self-checksResources MentionedThe Leadership Operating System by Jürgen Dauk: https://www.amazon.com/Leadership-Operating-System-Accelerating-Dominating-ebook/dp/B0GX2TNS92Leadership Operating System website: https://theleadership-os.comDesign thinkingThe Innovator’s Dilemma by Clayton ChristensenListen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#604 AI Can Generate Expertise. It Still Can’t Generate Judgment | Dan Pratl, Founder & CEO, Quadron
3w ago00:40:44Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Dan Pratl, Founder and CEO of Quadron. Dan is building infrastructure around trust, credibility, reputation, and human judgment in a world where AI can generate expert-looking work at near-zero cost.The conversation reframes one of the most common assumptions about AI. The scarcity is no longer knowledge creation. The scarcity is verification, judgment, and the ability to demonstrate that a person stands behind a claim. Rather than treating AI as a replacement for expertise, Dan argues that AI increases the value of trusted human judgment.If you are building, investing in, operating, or leading in AI, enterprise software, digital infrastructure, or knowledge-intensive businesses, this conversation provides a framework for thinking about trust, reputation, and value creation in an AI-driven economy.About the GuestDan Pratl is the Founder and CEO of Quadron, a company focused on creating infrastructure for trust, credibility, reputation, and programmable incentives in the AI era.His background spans regulation, open source software, crowdfunding, decentralized finance, and crypto. Through those experiences, he developed a thesis that human expertise, judgment, and credibility should become measurable, portable, and economically valuable assets.His work focuses on solving a problem that becomes increasingly important as AI-generated content becomes abundant: determining who stands behind information and why that credibility should matter.LinkedIn: https://www.linkedin.com/in/danpratl/Website:https://quadron.tech/Personal Site:https://pratl.meKey Takeaways• AI has made knowledge generation abundant, but trust remains scarce.• The value of expertise increasingly comes from judgment rather than content creation.• Traditional credentials and social proof systems are losing effectiveness.• Credibility needs to become portable rather than tied to individual platforms.• Verification must become a byproduct of human ambition and incentives.• Human expertise is an evolving asset that compounds over time.• AI agents can execute tasks, but humans still define what good looks like.• Organizations that capture and reward human judgment will outperform those that only optimize automation.What You Will Learn• Why AI-generated expertise does not eliminate the value of human judgment.• How credibility may evolve into a measurable and portable asset.• The limitations of resumes, endorsements, and traditional reputation systems.• How programmable incentives can encourage verification and trust.• What a credibility wallet could look like in practice.• Why AI agents still depend on humans to define outcomes and quality.• How organizations can preserve and scale expertise in an AI-first environment.Episode Highlights00:00 — AI Makes Trust More Valuable Than Knowledge05:00 — Knowledge Becomes Abundant, Verification Becomes Critical08:00 — Why Judgment Outlasts AI Generated Expertise11:00 — The Case for a Portable Credibility Wallet14:00 — Quantifying Reputation Beyond Social Proof16:00 — Expertise Compounds Through Iteration18:00 — Turning Judgment Into an Economic Asset21:00 — Investing in Yourself as a Market25:00 — Verification Must Reward Participation30:00 — AI Agents Need Humans To Define Good33:00 — Companies That Ignore Human Judgment Fall Behind35:00 — Building a New Category Around Trust InfrastructureResources Mentioned• MCP (Model Context Protocol)• Skills.md• Red Hat• SEC (U.S. Securities and Exchange Commission)• CFTC (Commodity Futures Trading Commission)Listen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, venture capital, AI, cybersecurity, and enterprise technology.
Transcribe →#603 Startups Scale Too Early. The Basics Are Still Broken | Raphael Peyret
3w ago00:59:43Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Raphael Peyret, Founder and Principal Advisor at SHA/RP. Raphael brings experience across product, cybersecurity, Google, and startup execution from MVP to acquisition. The main tension is clear: companies keep chasing scale before the basics are working.The conversation reframes AI security, startup growth, product management, and GTM as the same sequencing problem. AI-native threats matter, but unpatched systems, weak credentials, poor MFA adoption, unclear positioning, premature sales hiring, and feature overload still break companies first. Raphael argues that founders need defensible security, repeatable sales, and product discipline before they scale people, spend, or complexity.If you are building, investing in, or leading early-stage enterprise technology, cybersecurity, AI, or SaaS companies, this conversation gives a practical way to separate progress from motion.About the GuestRaphael Peyret is the Founder and Principal Advisor at SHA/RP, where he works with startups as an independent advisor and fractional executive across product management and cybersecurity.His background includes Google and a VP of Product role at Harangi Cybersecurity, a Singapore-based cybersecurity startup that moved from MVP through fundraising, acquisition, and integration into Bitdefender.Raphael frames startup execution through the lens of risk, product discipline, and sequencing, which makes him well placed to discuss where founders and security leaders usually move too early.LinkedIn: https://www.linkedin.com/in/rpeyret/Website: https://sha-rp.comKey TakeawaysAI threats get attention, but basic security failures still cause most breaches.Startups need defensible security, not enterprise-grade security theatre.Cybersecurity should help startups move faster without creating reckless exposure.Founders often hire sales before they understand how their product sells.A salesperson cannot fix unclear positioning or unfinished customer pain.Product teams fail when they add features before solving the core problem.Founder bottlenecks appear when decisions stay personal instead of becoming systems.Motion becomes progress only when each step proves a specific assumption.What You Will LearnThe difference between AI security headlines and the breach risks most companies actually face.How startups can define good enough security without copying enterprise playbooks.Why basic hygiene such as MFA, SSO, and credential management still matters most.When hiring sales too early creates more confusion than revenue.How product management helps founders stop becoming the bottleneck.Why feature expansion can hide weak product-market understanding.What separates motion from progress in founder execution.Episode Highlights00:00 — Raphael Peyret connects cybersecurity with startup execution02:00 — AI threats distract from basic security failures05:00 — Security teams still struggle to speak business language09:00 — Startups need defensible security, not overbuilt controls15:30 — Security diagnostics expose the risks founders miss18:00 — MFA and SSO still form the security base20:30 — Good enough security helps startups keep moving24:30 — AI can reduce friction before attacks begin27:00 — Startups hire sales before sales is repeatable31:00 — Marketing cannot fix unclear positioning35:00 — Product teams add features before solving pain40:30 — Founders need systems before they can scale46:30 — Fractional leadership bridges the early expertise gap49:30 — Motion and progress are not the same thing56:30 — Founders need sequencing across every functionListen NowAvailable on all major podcast platforms and YouTube.Follow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#602 AI Can Set Meetings. It Still Can’t Build Trust | Alex Grant
4w ago00:54:04Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Alex Grant, SVP of Sales at North. Alex brings more than 16 years of experience building sales teams across fintech, payments technology, and SaaS. The conversation centers on a hard tension: AI can create more opportunities, but it still cannot create trust by itself.The conversation reframes AI in sales and fintech as an execution problem rather than a technology topic. Alex explains why companies are using AI to move faster, secure payment systems, and generate more qualified opportunities, while also showing why complex software still needs a human seller who can translate risk, value, and trust for the buyer.If you are leading revenue, building fintech products, investing in AI-enabled software, or selling complex enterprise technology, this conversation shows where AI can accelerate the system and where human judgment still carries the deal.About the GuestAlex Grant is the SVP of Sales at North, where he is building the company’s first coast-to-coast W-2 outside sales channel. Before joining North, Alex spent more than 16 years building sales teams in fintech, payments technology, software, and SaaS, including time at a Fortune 500 company.At North, Alex focuses on building full-time sales teams that can sell payment technology, AI-supported security, and software solutions with a structured career path, training model, and field-led culture. His perspective is grounded in the operational reality of selling fintech and payments technology to businesses of different sizes.LinkedIn: https://www.linkedin.com/in/ralexgrant/Website: https://north.comKey TakeawaysAI can generate meetings, but it cannot replace trust in complex technology sales.Buyers want AI, but many still struggle to define what they actually need.Payment security is one of the clearest practical use cases for AI in fintech.Smaller businesses often underestimate security risk until they become easier targets.AI lowers the cost of testing new software ideas before committing years of development.Automated SDR workflows will pressure traditional appointment-setting models.Human sellers still matter when buyers need confidence before signing large contracts.Sales teams perform better when leadership gives field teams real access and voice.What You Will LearnHow AI is changing prospecting and appointment setting in fintech sales.Why payment security is becoming a stronger AI use case than generic productivity.The reason smaller merchants often misunderstand their exposure to fraud and breaches.How buyers talk about AI when they know they need it but cannot define the purchase.Why complex software still needs human interpretation during the sales process.When W-2 sales teams create more control than 1099 agent-led distribution.What sales leaders can do to keep field teams engaged, heard, and useful.Episode Highlights00:00: AI sales needs a human interpreter05:00: Payment security becomes the practical AI case07:30: Small businesses misread their breach exposure11:30: Buyers want AI before defining the need14:00: AI lowers software development risk17:00: Automated SDRs pressure old appointment models19:00: Trust still decides large software purchases28:30: North shifts toward a W-2 sales model36:00: Salespeople need access, not just incentives45:30: Door-to-door selling still creates human trust50:00: AI turns ideas into working prototypes fasterResources MentionedNorth: https://north.comW-2 sales model: referenced as North’s full-time sales channel structure1099 agent model: referenced as the traditional distribution model in payments technologyListen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#601 The AI Bottleneck Is No Longer GPUs. It’s Energy and Memory | Eugene Cheah
May 2500:45:37Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Eugene Cheah, CEO of Featherless AI. The AI bottleneck is no longer just GPU access. Power, memory, inference cost, and model reliability are becoming the real constraints.Eugene reframes the AI infrastructure debate away from a simple race for bigger models and more chips. The conversation connects energy capacity, HBM shortages, open source model adoption, linear attention architectures, and the enterprise need for predictable AI systems. It also challenges the assumption that the best AI strategy is always to use the largest available model.If you are building, investing in, or operating AI infrastructure, this conversation gives a clearer view of where AI economics, hardware constraints, and production reliability are heading.About the GuestEugene Cheah is the CEO of Featherless AI, an AI startup making open source AI models accessible through a single platform.Featherless AI started from AI research and optimization work around RWKV architecture, with a focus on reducing inference cost and making AI models more accessible. Eugene’s work sits at the intersection of open source AI, model efficiency, GPU infrastructure, HBM constraints, and inference optimization.He is well positioned to frame this shift because Featherless AI works directly on the infrastructure layer between developers, open models, and production inference.LinkedIn: https://www.linkedin.com/in/eugene-cheah-a47791126/Website: https://featherless.aiKey TakeawaysAI infrastructure constraints are shifting from GPU access to power, memory, and inference efficiency.HBM scarcity becomes more serious as models and context windows continue to grow.Bigger models do not solve the enterprise problem of reliable execution.Open source models are becoming strong enough to replace many closed model use cases.Fine-tuned smaller models can outperform frontier models on narrow enterprise tasks.Nvidia’s moat weakens when developers can move workloads across more hardware choices.Linear attention architectures matter because quadratic memory scaling is economically unsustainable.Enterprises value model control when closed providers change, deprecate, or restrict models too often.What You Will LearnThe real infrastructure bottlenecks behind AI deployment beyond GPU availability.How HBM pressure affects model size, context length, and inference economics.Why energy capacity can delay AI infrastructure even when chips are already available.How open source models are changing enterprise AI adoption and deployment control.Why smaller fine-tuned models can beat larger models on specific production tasks.When linear attention architectures reduce memory demand compared with transformer attention.What hardware choice, model portability, and local inference mean for AI infrastructure strategy.Episode Highlights00:00 — AI infrastructure moves beyond the GPU race03:30 — Nvidia, AMD, and Huawei follow different hardware strategies07:30 — Power becomes the first AI infrastructure bottleneck08:30 — HBM pressure exposes the memory constraint12:00 — AI follows the same pluralism as databases15:00 — Developers start with big models, then specialize18:30 — Transformer memory scaling becomes an economic problem23:30 — Hardware choice starts weakening platform lock-in29:30 — Reliability matters more than raw intelligence36:00 — Open source gives enterprises model control41:30 — Small models can now build real applicationsResources MentionedFeatherless AI: https://featherless.aiRWKV architecture: AI architecture referenced by Eugene as part of Featherless AI’s research backgroundListen NowAvailable on all major podcast platforms and YouTube.Connect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →#600 AI Reliability Is a Business Risk. Not Just an Engineering Problem | Helen Gu
May 2200:37:26Tap to summarizeIn this episode of The CTO Show with Mehmet, Mehmet sits down with Helen Gu, Founder and CEO of InsightFinder AI. Helen brings decades of research in distributed system reliability, anomaly detection, and AI-driven operations. The conversation focuses on why AI reliability is becoming a business risk, not just an engineering issue.The conversation reframes AI observability as a production control layer for enterprises deploying AI agents. Helen explains why traditional DevOps and SRE practices are not enough when systems are probabilistic, model behavior changes, data shifts, prompts evolve, and agents begin taking actions across workflows.If you are building, investing in, operating, or leading AI systems inside enterprise environments, this conversation gives you a practical frame for reliability, drift, runtime monitoring, and accountability.About the GuestHelen Gu is the Founder and CEO of InsightFinder AI, and a professor at North Carolina State University. InsightFinder AI was founded from her research in distributed system reliability using AI technology.Helen has worked on anomaly detection, prediction, diagnosis, and system reliability since the late 1990s. She also spent a sabbatical year at Google evaluating anomaly detection algorithms, which later helped shape the foundation for InsightFinder AI.LinkedIn: https://www.linkedin.com/in/helen-gu-b1aa42b6/Website: https://insightfinder.com/Key TakeawaysAI systems can fail silently while still returning confident answers.AI reliability is becoming a business risk, not only an engineering concern.Multi-agent systems can spread upstream mistakes across business workflows quickly.Traditional SRE practices do not fully cover model behavior, prompts, and data drift.Runtime monitoring matters more once AI moves from sandbox testing to production.Observability alone is not enough without diagnosis, recommendations, and remediation.Model drift can change business outcomes even when infrastructure appears healthy.Human review shifts from doing work to supervising AI decisions and guardrails.What You Will LearnWhy probabilistic AI systems require different reliability practices than software systems.How model drift and data drift change production behavior over time.What silent AI failure looks like inside enterprise workflows.The reason sandbox testing misses real production AI failure cases.How runtime monitoring helps detect hallucinations, bias, leakage, and accuracy issues.Why AI observability must connect infrastructure, data, prompts, models, and business outcomes.What leadership teams need to consider before AI agents begin taking actions.Episode Highlights00:00 — Helen Gu frames AI reliability from research02:30 — AI systems answer confidently even when wrong04:30 — SRE lessons do not fully transfer to AI07:00 — AI reliability needs fine-grained runtime metrics08:30 — Silent failure creates hidden business damage10:00 — Multi-agent mistakes propagate faster than humans12:00 — Model drift changes outcomes without warning15:00 — Sandboxes miss production AI behavior18:00 — Observability must become actionable control21:30 — AI reliability becomes a leadership responsibility24:30 — AI Labs test prompts, models, and datasets28:30 — AI agents become part of enterprise workflows31:30 — Responsible AI starts with accepting failure riskListen NowAvailable on all major podcast platforms and YouTubeConnect with the ShowFollow The CTO Show with Mehmet for more conversations at the intersection of technology, startups, and venture capital.
Transcribe →