https://www.gtmaipodcast.comYour AI conversations are sitting in someone else's vault. In this episode, 20-year GTM operator John Williams shows exactly how he took his back: archiving every chat locally, running AI models offline on his own laptop, and setting hard guardrails on what his agents can buy and agree to without him.This is the GTM and AI Podcast, where real operators show the receipts. One rule in this kitchen: you actually have to cook.WHAT JOHN SHOWS LIVE ON SCREEN:• Chat Archive: a free, open-source browser extension that exports any AI conversation (Claude, ChatGPT, Gemini, Groq, Perplexity) to JSON or markdown, with zero outbound calls. Nothing leaves your machine.• Exporting a full Claude conversation and continuing it inside Groq with full context intact• Why he runs local models with Ollama, and how open-source models let you switch models mid-conversation without losing context• Agent Commerce: an open spec for what your AI agents can spend and what terms they can accept on your behalf• The AI Acceptable Use Policy: an open-source starting point so shadow AI doesn't run your company• How to vet any open-source AI tool before you trust it (the one question: would your security director approve?)TIMESTAMPS:00:00 - Welcome to the kitchen: the one rule of this podcast01:00 - Who is John Williams? 20 years in GTM, 5 as an independent operator02:30 - Why your GitHub repo is the new resume04:00 - The AI Acceptable Use Policy: fixing the shadow AI problem05:30 - Agent Commerce: spending limits and guardrails for your AI agents09:00 - Chat Archive: why owning your conversation history matters14:00 - Building a digital twin from a year of AI conversations16:00 - Local models 101: moving past being a "prompt jockey"19:00 - GPU brownouts and token authority: why local inference is your backup plan22:00 - LIVE DEMO: exporting a Claude conversation locally25:00 - Porting full context from Claude into Groq, zero loss28:00 - Token economics: the W-2 cost didn't disappear, it moved31:00 - Data portability use cases: audits, regulated industries, federated intelligence36:00 - OpenClaw: the power and the security risks of autonomous agents39:30 - Where to find John + why he'd apply for his next job in publicFIND JOHN WILLIAMS:GitHub: github.com/fxops-aiHugging Face: huggingface.co (johnwilliamsatl)If this episode saved you from losing a year of AI conversations, subscribe and come cook with us next week.#GTMAI #AIAgents #LocalAI #DataOwnership #Ollama #GoToMarket

https://www.gtmaipodcast.com

He stopped hiring for RevOps tasks and started building Perplexity skills instead. Here's exactly how.

In this episode of the GTM AI Podcast, Coach K sits down with Nate Follen, Head of Enterprise Ops and Systems at Perplexity, to walk through the three AI skills his team actually runs every day and every week. Nate helped scale RevOps at Ramp, advised teams at Momentum, and now leads the internal go-to-market systems engine at Perplexity. He doesn't talk theory. He shares his screen and shows the real workflows.

Key topics covered:

• Why Nate now asks "do I have to hire for this?" before adding headcount, and where the answer is still yes (hint: the Salesforce admin role isn't dead)

• The Voice of Customer dashboard he built with two prompts, an API key, and live-refreshing call transcripts, that tells the product team not just what customers said, but what to do about it

• The weekly RevOps deck-prep skill that pings the sales team in Slack, pulls live data from Snowflake via Claude Code, checks Linear, and assembles a deck that used to take an hour

• The CRM hygiene system that cleans account ownership and Salesforce hierarchies nightly, builds Polytomic data models without logging into the tool, and DMs Nate every error with a severity score and a fix

• Why orchestration across 400+ connectors beats single-model lock-in, and where Perplexity sits next to Momentum and Salesforce instead of replacing them

• The mindset shift: from "find the two things to focus on" to running dozens of projects in parallel with a team of agents

If you lead revenue operations, enablement, sales, or marketing, this is the tactical breakdown of what AI-run GTM systems actually look like inside one of the fastest-growing AI companies in the world.

Resources mentioned:

• Nate Follen on LinkedIn: https://www.linkedin.com/in/follen/

• Perplexity: https://www.perplexity.ai

• Momentum.io (call recording and transcript analysis)

• Salesforce, Slack, Linear, Snowflake, Claude Code, Polytomic, Hightouch

• GTM AI Podcast & Newsletter: www.gtmaipodcast.com

Get the free lead magnet from this episode, The Perplexity RevOps Skill Stack, with the exact prompts and build steps: www.gtmaipodcast.com

https://www.gtmaipodcast.comhttps://www.fluint.comIn this episode of the GTM AI Podcast, I sit down with Nate and John, co-founders of Fluint, for a kitchen-deep look at how they built Ollie, an omni-agent that captures tacit sales knowledge and surfaces it across an entire enterprise sales org. We cover:The tacit knowledge problemWhy a small pocket of reps wins 50-60% of pipeline while everyone else sits at 17-19%, why "clone your top rep" has been said for 15 years and never delivered, and what happens to your $30M deal architect when they take a new job two weeks from now.The implicit signal exampleA real story of two reps, two POC readouts, two procurement follow-ups (one at 9pm Friday, one Wednesday during business hours), and why top reps will negotiate completely differently on the same data while average reps miss the signal entirely.The photo vs. video architectureWhy most AI tools treat data as a snapshot (LLM context = one photo), and why Fluint's event-driven, time-series architecture treats it as a video. The 10-Second Tom analogy from Fifty First Dates that explains why LLMs alone cannot solve this problem.The ML + LLM stackJohn walks through the architectural decision: ML for pattern recognition and judgment layer, LLM for human-to-human interaction. "Using the right tool for the right job." This is the most underrated decision in enterprise AI right now.Ollie's omni-agent designWhy one AI teammate beats 130 task-specific agents. The 75% of users who gender their AI. The trust dynamics that make a sales rep follow an agent's advice when it runs counter to the playbook.The racehorse modelHow Fluint runs a global baseline model and a customer-specific model in parallel, evaluates them nightly, and promotes the winner. Continuous evaluation as the moat.The "data is a product of people" answerJohn on why perfect data is a logical fallacy and what to do instead. The single line that changes how you approach AI readiness.Real outcomes+$28K added to ACV per team per year. 32 days off the median sales cycle. The maturity curve from Q1 (win existing deals with less discount) through year-end (win deals you would have lost).GUESTSNate, Co-founder & CEO, Fluint (the "second brain")Repeat enterprise sales leader and repeat founder. Built Fluint from a problem he could not solve as a sales leader. Author of two books on tacit knowledge and executive sound-bite communication.John, Co-founder & CTO, Fluint (the "first brain")Technical co-founder. Builds the systems that turn Nate's crazy ideas into shipping product. Specialty: event-driven architectures and ML-as-judgment-layer for enterprise sales.LinkedIn Nate:https://www.linkedin.com/in/natenasralla/Linkedin John:https://www.linkedin.com/in/jon-crawley-3797a8100/Blog: fluint.io/blogJohn's recent technical guide: building enterprise AI agents (just published on the Fluint blog)GitHub repo (DIY resources for time-series-data agent architecture): linked from blog