https://www.gtmaipodcast.comMark Fershteyn calls himself "a dumb sales guy with too many ideas." He ran sales at App Academy, founded Recapped, raised $8M, built a 26-person team, reached seven figures, and eventually shut it down and returned money to investors.Then AI changed the math. For his whole career, building an idea meant raising money and hiring engineers. Now he builds it himself.In this episode, Mark shares his screen and walks through:How to keep up with AI without burning out (filter for step changes)Why he starts almost every session in plan modeThe "one agent per task" rule and his 6-agent content pipelineA skill built from 300 LinkedIn posts that interviews him, drafts, and scores 5 hooksHow he structures folders and a personal wiki so AI can find everythingZealos: 5 calendars, every inbox, every DM, website visitor identification, and sequences in one appThe revenue: a $500K run rate in 6 months, solo, part-timeBuild vs. buy vs. hire, and where the software market goes nextIf you are not technical and you think this stuff is out of reach, watch this one.Chapters00:00 Intro: Mark's second time on the show01:11 From VP of Sales to an $8M-raised founder02:10 Founder tip: don't create a new category02:40 Why AI changed everything for a non-technical builder04:03 How to keep up without drowning05:14 Filter for step changes, then build something05:58 AI as playtime07:00 What you'll learn: skills and a real build08:15 Claude Code in the terminal vs. the app09:24 Use AI to use AI: start in plan mode10:26 One agent per task11:31 The prompt that builds your content agent system12:20 Coach K's rule: four rounds of feedback13:07 Treat your agent like an intern13:30 The 300-post LinkedIn skill14:08 Make every skill improve itself14:49 Building an AI CMO (plus a video editing tool tip)16:00 Folders, GitHub, and the /clients folder17:08 Hierarchy vs. graph: building a personal wiki18:58 Zealos: the super app20:39 From 20 tabs to one view21:58 Identifying website visitors and enrolling them in sequences23:05 From personal tool to product24:20 The revenue impact25:16 4 years and 26 people vs. 6 months solo26:00 UI vs. chat: everyone gets a Jarvis29:20 Build vs. buy vs. hire30:41 SMBs build, enterprises buy, the middle gets squeezed32:00 The build-happy pendulum32:37 Where to find MarkLinksMark on LinkedIn: https://www.linkedin.com/in/markfer/Mark's links: markfersh.comMastering AI: masteringai.ioZealos: zealos.ioPalmier (video editing via Claude Code, mentioned by Coach): palmier.ioRemotion (video editing, mentioned by Mark): remotion.devFree Content Engine skill: https://www.therevenueaireport.com/agent-skills/content-engine-suite

https://www.gtmaipodcast.com

Eli Portnoy has spent 15 years in AI across three companies. ThinkNear was acquired by Telenav, Sense360 by Medallia, and BackEngine.ai is the one he says he is not planning to sell. The premise of BackEngine: AI needs context, most company data is siloed and inaccessible, so BackEngine structures, permissions, and indexes private company data so every AI instance can use it.

On this episode Eli shares his screen and walks through how BackEngine runs 90% of its go-to-market inside Claude.

What you'll hear:

• The connector stack. BackEngine, Fireflies, Gmail, Calendar, Granola, HubSpot, Notion, Slack, Superhuman, Zoom, plus a few vibe-coded MCP servers (website stats). The SaaS tools didn't go away; they moved inside Claude.
• The scheduled jobs that survived. Founder Sales Daily Pulse (4pm outreach scan), end-of-day follow-up audit from transcripts, product-market fit review every 20 prospect calls, and weekly deal/customer health change alerts.
• Why a direct CRM connector misses deals. Context windows force the model to sample. Eli's estimate: about 30% of the relevant data gets read. Joining systems, building an index, and permissioning at the data layer fixes it.
• How to measure AI ROI when it's hard to attribute. The "30% smarter pill" thought experiment, and why scheduling turns AI from single-player into multiplayer mode.
• Three mistakes. Capabilities weren't ready (jobs needed a computer on and permissions every run), running 200 jobs nobody read, and treating adoption as a technology problem instead of change management.
• Two change-management rules. Fewer jobs with owners, and no job without a workflow attached.
• Artifacts as dashboards. Rebuilding the siloed SaaS views (customer health by tier, account health, case studies) as shareable Claude artifacts.
• Prompt vs. project vs. skill vs. plugin. Eli's 20-second taxonomy.
• State of the market. 5-10% of employees in most companies are power users; 90% use AI lightly or for emails. Causes: fear, habit, immature tooling, no formula.
• Advice for individuals. Play with it, give it context like you would a new hire, review every output, tell it to be concise.
• The next 12 months. Chat becomes the central work surface; expect non-model-provider layers (Manus, Instinct) to compete on the machinery around the model; voice and other interfaces are coming. Eli's analogy: early TV filmed radio booths.

https://www.gtmaipodcast.comJoel Horwitz ran product-led growth for all of IBM, led growth at Sourcegraph, and now runs Synter, an ad platform operated by AI agents. In June he ran a LinkedIn campaign that came back at $1,800 per lead. In this episode he shares his screen and shows exactly why that happens, why it took him six months to figure out, and the workflow he uses now instead.

You'll see the one-prompt "demand capture" workflow that turns your zero-click organic keywords into exact-match paid campaigns, the LinkedIn API tier problem that quietly sends your budget to Walmart employees and BDRs, the organic-first playbook that took Synter from 0 to 2,500 followers with no ad spend, how to scope permissions for agents that can spend money, and a warning about malicious skill files.

Real screen share. Real numbers. No theory.

🎙️ GTM AI PODCAST More episodes, the newsletter, and the community: https://gtmaipodcast.com

👤 GUEST: JOEL HORWITZ Synter: https://syntermedia.ai/ Joel's DMs are open on X