The A to Z of AI

In this episode, Anthony Thomas and Zac Henderson are joined by Avi Gesser, partner and co-chair of the Data Strategy and Security Group at Debevoise & Plimpton, where he also leads the firm's AI practice. Avi explains how the firm builds purpose-specific custom GPTs, including Patty, a cautious research and writing assistant credited as a coauthor on firm blog posts, and Sergio, a brainstorming counterpart with the opposite instructions. He also walks through STAAR, a subscription portal that gives clients direct, searchable access to the firm's genericized AI advisory work.The conversation covers what happens to a law firm's model when much of its work becomes repeatable. Avi argues that firms should use AI to scale their expertise, that precise feedback improves model output more than most people expect, and that clients still want a human conversation even when the answer is already in front of them. He is also candid about the limits: agent workflows are hard to audit, AI writing lacks a voice, and training junior lawyers will take more deliberate effort.Key TakeawaysGiving each assistant a clear purpose and instructions changes results more than the underlying model doesNaming assistants encourages people to give feedback and iterate, as they would with an associateReaching roughly 50% regular adoption took dedicated staff, internal demos, and sustained effortSTAAR replaces hourly billing for repeatable AI work with a subscription that funds new shared contentAbout 80% of clients still call after STAAR returns the exact answer they neededAvi prefers chained tasks with human checks over agent orchestration, because errors are easier to traceAssociates should use AI for ideas and arguments, and write client-facing work from scratchTo see what current tools can do, Avi suggests drafting lyrics about a friend with a chatbot and generating a custom song in SunoAbout the PodcastThe A to Z of AI is hosted by Anthony Thomas and Zac Henderson and powered by The Suite. The show explores real-world AI use cases across industries, focusing on practical workflows, tools, and lessons learned from operators using AI in their daily work.

Alex Triplett, COO of You.com, joins Anthony Thomas and Zac Henderson for a practical look at how AI adoption actually gets funded, structured, and owned inside a company. You.com supplies web search infrastructure to AI agents and LLMs, and Alex owns go to market, G&A, and R&D across the business. He walks through his own timeline from manual front office work in 2022 to the agent-driven workflows his teams run today.The core of the conversation is a challenge to how most companies approach AI budgets. Alex argues that organizations demanding precise ROI measurement before adoption never had that measurement for human productivity in the first place, and that spend caps do more damage to experimentation than they save in cost. The discussion then gets specific about who owns AI workflows in a go to market org, where human judgment still has to sit, and why he routes work to open weight models wherever quality allows.Key TakeawaysSpend caps on AI tools restrict experimentation far more than they control costJudge AI spend against the fully loaded cost of the people it makes fasterAdoption needs executive buy in and education at every level, without committees or formal AI programsLet teams over-experiment first, then standardize templates and route cheaper models where quality allowsAI can carry 85 to 90 percent of an outbound sequence, with humans owning the final polishGive AI workflows clear owners: systems engineering for the stack, rev ops for consistency, an executive for reviewOpen weight models deliver comparable output at lower token cost and are worth routing ancillary work toAbout the PodcastThe A to Z of AI is hosted by Anthony Thomas and Zac Henderson and powered by The Suite. The show explores real-world AI use cases across industries, focusing on practical workflows, tools, and lessons learned from operators using AI in their daily work. 

Mercedes Bent, founder of Premise VC and formerly a partner at Lightspeed, joins Anthony Thomas and Zac Henderson for the season's first guest conversation. She walks through the stack she uses across the five jobs of venture, sourcing, evaluating, winning, portfolio support, and exits, then demos the agents her team built to run an in-person community of roughly 450 AI engineers and researchers.The through line is her argument that the obvious first use of AI, more volume at lower cost, is usually the wrong one. Getting to something genuinely useful takes second and third order thinking about what the work is actually for. That is why she points automation at logistics and coordination while keeping investment judgment, live conversations, and the parts of the job she enjoys in human hands.Key TakeawaysMost first-order AI use is just faster spam, and better applications require rethinking the purpose of the workAgents are best pointed at logistics and coordination, where the hours disappear without adding valueA network only creates value when members can see who is actually in itModel quality and pricing are being tested in real time, and heavy users notice degradation within daysFounder quality still drives most of the underwriting decision, so research what a specific investor needs before pitchingProtect the work you would be embarrassed to be unable to defend in a room, starting with your own reasoningMercedes recommended trying Super Me, the digital twin product her firm invested in, where her own agent is trained on her public content plus her private meeting notes and has helped founders practice pitches and prepare for interviewsAbout the PodcastThe A to Z of AI is hosted by Anthony Thomas and Zac Henderson and powered by The Suite. The show explores real-world AI use cases across industries, focusing on practical workflows, tools, and lessons learned from operators using AI in their daily work.

Season three opens with Anthony Thomas and Zac Henderson on their own, taking stock of where AI adoption actually stands after two seasons of guest conversations. Zac has moved from technology general counsel to teaching law at SMU, and the discussion covers the BNY CEO who rates himself a nine out of ten on AI while admitting he uses a fraction of what is available, the cost pressure pushing companies toward open models, and what watermarking under the EU AI Act means for work product.The through line is restraint. Both hosts describe pulling back over the past quarter: Zac shut down most of his agent workflows after concluding he spent more time managing them than he gained from them, and Anthony has yet to get a proactive chief of staff agent working. Their argument is that AI is expanding outward faster than it is advancing forward, which makes deciding what to skip as important as deciding what to learn.Key TakeawaysAI is expanding outward faster than it is advancing, so narrow focus beats broad coverageBNY's CEO rates himself a nine out of ten on AI and still estimates he uses 5% of what is possibleStanding daily briefs and weekly call review bots are the highest-leverage personal setups discussedRouting simple tasks to cheaper open models cuts cost but adds governance and evaluation burdenWatermarking has a second edge as a licensing lever, not only a provenance signalLaw schools are moving toward oral defense of written work as the real test of judgmentSpinning up a new agent still costs 5 to 10 hours on a medium task, often more than it returnsAbout the PodcastThe A to Z of AI is hosted by Anthony Thomas and Zac Henderson and powered by The Suite. The show explores real-world AI use cases across industries, focusing on practical workflows, tools, and lessons learned from operators using AI in their daily work.

In the Season 2 finale, Anthony Thomas and Zac Henderson are joined by Ellen Li, CFO at AI Fund, the venture studio founded by Andrew Ng that builds AI-native startups from the ground up rather than investing in deal flow. Ellen walks through how her day-to-day has shifted from formatting spreadsheets and manual research to reviewing AI output as she would a capable team member. She shares concrete builds: a folder-scanning script that has saved her team hours a week for over a year, and an investor sourcing pipeline cobbled together with Claude's MCP connectors across Apollo, HubSpot, Gmail, and more.The conversation is grounded by a distinction worth holding onto: automation a non-engineer can do tends to live in the task, not the full workflow. Ellen is candid about the limits, including a portfolio-monitoring tool she abandoned at two in the morning after overestimating what AI-assisted coding could do for her. Her larger point is that the constraint has moved. Coding is cheap now, so the scarce skill is judgment about what to build in the first place, and where buying still beats building.Key TakeawaysNon-engineers can reliably automate individual tasks, but full workflows still need real engineering.AI is dependable on general financial analysis and weak on niche knowledge that lives outside its training data.Knowing what to build next is now more valuable than the ability to build it.Build versus buy still matters, because someone has to maintain, debug, and own what you build.Start by automating one workflow you genuinely hate, then expand slowly.Token maxing is a vanity metric; ROI and a clear North Star are the real measures.Treat AI output like a strong team member's draft that you still review and stand behind.About the PodcastThe A to Z of AI is hosted by Anthony Thomas and Zac Henderson and powered by The Suite. The show explores real-world AI use cases across industries, focusing on practical workflows, tools, and lessons learned from operators using AI in their daily work.