Fringe Lines

Quinn and Doom discuss how high-agency “AI-pilled” users are achieving outsized productivity gains, driving fear of a “SaaS apocalypse” as people can build custom tools—like a real-time ClickHouse dashboard—directly with Claude Live Artifacts instead of buying SaaS. They review headlines including massive AI infrastructure spending, Anthropic/OpenAI moving into services via “forward deployed” engineers, Thoma Bravo’s Medallia LBO failure that wiped out equity, and Google’s three-pillar AI paywall strategy. They debate why enterprises lag in AI adoption due to data access, silos, connectors, and change management, and how cost and token usage may clash with existing BI tools. The episode then shifts to Andrew Karpathy’s “second brain” concept using Obsidian vaults and Claude Code to curate notes/web clippings into a personal wiki and knowledge graph for faster retrieval and context-aware work.
 
00:00 AI 100X Productivity
00:50 Live Artifact Dashboard Demo
02:29 Dashboards for Any Business
03:22 Karpathy Second Brain Tease
03:37 Hawaii Small Talk Break
04:39 AI Industry Headlines Roundup
07:43 Services and Adoption Gap
15:13 Medallia LBO Breakdown
19:23 SaaS Churn and Bad Tools
22:32 Headless SaaS and APIs
25:01 Token Costs vs BI Tools
27:38 Personal CRM App Idea
29:30 Obsidian Second Brain Setup
33:56 Querying Your Personal Wiki
38:58 Agency and Better Prompts
42:10 Wrap Up and Weekend Plans

This week discuss the accelerating pace of AI releases and launches—highlighting OpenAI’s GPT-5.5 shipping six weeks after 5.4, Google Cloud’s 8th-gen TPU gains, Amazon/Anthropic AWS enablement, and rapid adoption of Claude tools like Cowork, Code, Desktop, and “live artifacts” that enable refreshable dashboards and iterative deck design. They debate major industry moves and rumors (SpaceX’s deal to acquire Cursor, Microsoft exploring Cursor), arguing distribution and developer workflow are key moats while xAI lacks enterprise route-to-market. They cover Anthropic’s revenue growth and enterprise mix, the open-source Kimmi 2.6 coding claims, and growing concerns about throttling, data-center power constraints, and rising token costs. The “SaaS apocalypse” theme centers on collapsing software pricing ceilings, churn driven by cheaper AI alternatives, skepticism about multi-year contracts, and outcome-based pricing as a survival strategy, plus uneven adoption across GTM and sales.
 
00:00 Intro
00:21 AI Launch Firehose
01:20 Live Artifacts Dashboards
03:18 Headlines GPT 5.5 More
03:53 Cursor Deal Moats
05:19 Anthropic Revenue Surge
09:02 Data Center Power Crunch
11:12 Dev Tool Workflow Wars
12:18 SaaS Pricing Ceiling Falls
13:34 Salesforce Headless APIs
16:51 Build Anything Solo
17:28 Layoffs And Cost Cutting
18:03 Outcome Based SaaS Pricing
18:15 Multi Year Contract Regret
19:09 AI Spend By Industry
19:46 Claude Hype In GTM
20:56 Perplexity Instagram GTM
23:21 Claude Desktop Breakdown
24:27 MCP Versus CLI Costs
25:19 Why Sales Adoption Lags
28:38 Using AI For Prospecting
30:02 Tools Claude Users Still Need
32:03 Productivity Versus Bandwidth
33:20 SaaS Margin Reality Check
35:04 Keeping Skills Sharp

Doom and Quinns discuss the widening gap between people who use AI daily and everyone else, describing the pace of new tools as a “fire hose.” They share hands-on experiences with Google Stitch for fast UI prototyping (including use with kids), and react to Anthropic’s new design-related Claude release and the timing of a CPO stepping off Figma’s board. They compare design and web-building workflows (Figma, Claude Canvas/Code, Paper Design, shadcn, Tailark, Framer, WordPress), and talk about using AI to generate landing pages, AB tests, and even a fantasy-football learning site. The conversation covers token anxiety, agent permissions and governance, the difficulty of code review at AI-generated velocity, SaaS pricing pressure, big-company AI spend, VC funding trends, and uncertainty about where to place product and career bets.
 
00:00 AI Power Gap
00:23 Weekly Catch Up
01:44 Mac Mini and Stitch
02:17 Stitch Demo Workflow
04:26 Claude Design News
06:57 Landing Pages Toolchain
09:35 Tool Pricing and Moats
10:49 Agency Economics Debate
14:23 AI Prospecting Arms Race
17:11 Code Review and Governance
18:29 AI FOMO vs Security Risk
19:30 SaaS Pricing Pressure
20:04 Budgets Without Revenue
21:28 VC Cash Flooding In
22:19 Token Pricing By Use Case
24:14 Career Bets In AI Era
25:49 This Week In AI Headlines
26:56 Google Gemini Sleeper
28:01 Coding Tools And Subscriptions
30:22 Building And Shipping Fast
30:45 Fantasy Football Site Idea
33:22 Publish It And Monetize
35:54 Token Spend And Price Hikes
37:13 Weekend Plans And Parenting
38:34 Wrap Up And Goodbye

Anthropic just hit a $30 billion revenue run rate — and that's not even the wildest story this week. In this episode of Fringe Lines, Quinn Devery and William Doom break down why OpenAI may be losing the enterprise race (and why it might not be their fault), what Anthropic's Claude Mythos and Project Glasswing mean for cybersecurity, and whether we're watching the early innings of a full-blown SaaSpocalypse.
Plus: Meta launches Muse Spark on a brand-new AI stack, AWS Trainium has a multi-billion dollar backlog, HubSpot is quietly winning the SaaS spend wars, and Quinn shares his CI/CD pipeline setup with Claude and Cowork.
🔥 Key topics this week:
Anthropic $30B ARR and 3.5 gigawatts of committed compute through 2031
Claude Mythos & Project Glasswing — zero-day exploit detection at a new level
OpenAI's Microsoft-exclusive distribution problem in the enterprise
Meta Superintelligence Labs ships Muse Spark on a rebuilt AI stack
a16z SaaSpocalypse charts: who's winning, who's losing, and why
Are tokens the new cloud? AI now commands 5-15% of enterprise tech budgets
Builder updates: CI/CD pipelines, Cowork scheduled tasks, and Opus 4.6
📊 Charts referenced: a16z "Charts of the Week: SaaSpocalypse Interrupted" 🔗 https://www.a16z.news/p/charts-of-the-week-saaspocalypse
🎙️ Fringe Lines is a weekly show where Quinn Devery and William Doom talk AI, enterprise tech, and the business of building in the age of agents — no hype, just real takes from people doing the work.
Subscribe and hit the bell. Leave us a comment if there's a topic you want us to cover next week.
00:00 Catching Up
01:15 AI Headlines Rundown
02:53 Anthropic Revenue Surge
02:59 AWS Trainium Backlog
07:04 OpenAI Enterprise Bind
12:16 Claude vs Chat Tools
15:38 Meta Muse Sparks
16:30 Tokens as New Cloud
21:50 SaaS Spend Winners
29:25 Builders and Workflows
34:10 Wrap Up
 
#AI #Enterprise #Anthropic #OpenAI #SaaS #CloudComputing #FringeLines

Doom and Quinn talk about a week of experimenting with Anthropic tools like Claude Code and Cowork, focusing on building “skills” using MCP integrations to query company data (ClickHouse/Metabase) via natural language with guardrails and anomaly flags, while noting the tedious human eval/review step. They discuss using Playwright MCP for visual website iteration and automating WordPress page creation via app passwords, including failures when scaling layouts and the need for tighter process/context. One speaker tests running Claude Code on a Raspberry Pi with remote access challenges and considers alternatives like Tailscale and Claude Code Cloud, plus plans to install Cowork on an upcoming Mac Mini. They also highlight real-world risks: Claude-generated customer support guidance was confidently wrong, and an Amazon bot-driven support experience was frustrating despite solid integrations. The episode closes with advice to lean in, stay curious, and build leverage despite AI’s current rough edges.
 
00:00 RSA Travel Catch Up
01:11 Building Skills With MCPs
02:09 ClickHouse Data Queries
05:15 Playwright MCP For QA
06:23 Claude As Socratic Coach
08:10 Raspberry Pi Remote Setup
11:37 AI Stack Vision
13:09 Relationships In Negotiations
15:55 AI Support Gone Wrong
21:12 WordPress Automation Lessons
26:58 Lean In To AI
28:38 Wrap Up And Next Week

The speakers discuss a hectic week leading into the RSA Conference in San Francisco, describing the logistical challenge of scheduling back-to-back executive meetings across locations and adjusting flights around leadership sessions. They compare RSA, Black Hat, and Amazon’s discontinued re:Inforce event, then shift to how they use sanctioned internal AI tools (including a suite with multiple model modes, Bedrock access, and Claude Code) and the limits of delegating high-stakes work to agents. They react to recent AI headlines (Anthropic features and partner investment, Jensen Huang’s comments on token spend, and OpenAI’s focus on core product) and debate whether AI changes product-building frameworks, emphasizing soft skills as increasingly important. They cover enterprise sales team-building advice, skepticism about premature scaling in crypto/AI startups, shallow SaaS moats, usage-based pricing trends, and propose a “one project/automation a day” challenge with documented results.
 
00:00 RSA Week Chaos
02:17 Internal AI Tools
03:13 Deal Cycle Orchestration
04:12 AI News Roundup
05:44 Jensen Token Productivity
07:51 Guardrails And Tradeoffs
10:12 Soft Skills New Hard
11:02 Enterprise Sales Team Build
12:13 Crypto Reality Check
15:10 AI Native Platform Risks
17:59 Low Level SaaS Killers
19:38 Bring Your Own Key
20:13 Shallow AI Moats
21:03 SaaS 80 20 Split
22:23 Platform Long Tail
23:10 Usage Based Pricing Shift
24:59 Partners and Architects
25:48 Rethinking Workflows
28:22 Humans Still Matter
29:37 One a Day Automation
31:16 Gemini Firebase Bakeoff
33:42 Enterprise Data Friction
35:54 App Building Reality Check
36:16 Dispatch and Token Limits
38:01 Weekly Challenge Wrap

The conversation covers how model providers subsidize compute costs, leading to token caps on max/enterprise plans and pushing some teams to shift backend usage to hyperscalers like Amazon Bedrock to avoid throttling, consolidate spend, and leverage cloud commitments, amid explosive token adoption (notably for coding). They discuss SaaS pricing moving toward seat-plus-usage models with AI credits as pass-through infrastructure, skepticism about outcome-based pricing at renewal, and the need to protect margins from power-user outliers. The hosts react to headlines including Microsoft’s Copilot cowork, Anthropic’s lean growth team, and Atlassian’s layoffs tied to workflow documentation, exploring automation’s impact on roles, unresolved coordination pain points like scheduling, and the gap between individual AI productivity and institutional value, arguing companies must redesign systems while new cognitive overhead emerges from managing agents.
 
00:00 Token Caps Tipping Point
02:26 SaaS Pricing Shift
03:35 Slack AI ROI Doubts
04:42 Vibe Coding Reality Check
08:29 Headlines Rapid Fire
09:23 Atlassian Layoffs Workflow
12:48 Sales Automation Limits
15:53 Copilot Cowork Take
16:41 Pricing Models 2025
22:06 SaaS Apocalypse Debate
23:29 Moats Speed And Data
25:42 Agents As Tutors
28:08 Institutional Vs Individual
31:00 Closing Optimism

Doom and Quinn discuss the Anthropic report estimating AI’s potential impact on jobs using synthetic and telemetry-based task data, debating whether it wrongly assumes a static economy without new job creation; they cite declining banking headcount and job postings, while noting historical counterexamples like ATMs and new industries such as SAP. They argue AI coding tools can boost productivity but still create technical debt and risk failures, referencing an incident where Claude Code deleted a production database, and conclude jobs will change and reward those who become “10x” operators with hands-on, at-scale experience. They pivot to SMBs as a major automation opportunity, describing a private-equity example automating hair salon “paper cuts,” and discuss vertical tools for practices like dentistry and the need for ongoing maintenance or “AI service managers.” They also cover paid AI communities/templates, the emergence of GTM engineering as a new role, and end with weekend plans.
 
00:00 Unemployment and AI report
00:18 How the impact model works
01:53 Static pie vs new jobs
02:53 Banking headcount warning signs
04:55 Coding copilots and tech debt
06:17 Diffusion everyone can build
09:10 AI mistakes and guardrails
10:07 Becoming a 10x operator
10:54 Job postings and messy orgs
13:00 Self checkout and ATM analogy
15:50 Programmers exposure debate
17:23 Target can finally build
19:34 Pivot to SMB automation
20:38 SMB opportunity visualization
22:28 Dental practice headcount impact
23:46 Keep Building Momentum
24:29 Maintenance Is the Real Work
24:56 AI Service Manager Idea
25:51 Automation Pricing Shock
26:57 From Demos to Real Value
28:34 Meetups and Early Days Vibes
30:26 Claude Skills Marketplace
32:29 Courses and Creator Grifts
34:35 GTM Engineering Emerges
36:45 DevOps and SRE Parallels
40:49 Weekend Plans and Pastrami
43:08 Wrap Up and Next Week

This week we share how we are using Claude Cowork and Claude Code, emphasizing how fast agent workflows are compressing research, content creation, and GTM tasks. One describes setting up OpenClaw on a Raspberry Pi, using Claude to generate a full local-service website structure with internal/external linking, and experimenting with skills for more repeatable automation. They demo Cowork connectors that turn a Gmail leads folder into an interactive dashboard with tiers and recommended actions, and a marketing plugin that generates a full warm-inbound email sequence with setup logic, suppression rules, A/B tests, and performance benchmarks. They also show rapid AI-generated market research and an auto-created slide deck on AI-native CRM categories, discuss Salesforce MCP integration and Agentforce, rising personal AI tool subscriptions, and the idea that as execution becomes “perfect,” differentiation and distribution matter most.
 
00:00 Weekend Check In
00:16 Claude Coworker Hype
01:07 OpenClaw Setup Struggles
02:13 AI Built Handyman Site
04:17 Agents Everywhere Soon
05:56 Skills And Recursive Automation
06:53 AI Video Presentation Demo
11:28 Gmail Leads Dashboard
15:17 Marketing Email Sequence
18:03 AI Research Deck Demo
20:24 Perfect Execution Era
21:51 Information Arbitrage Window
23:06 Beating Prompt Paralysis
24:54 Personal AI Tool Stack
26:26 Skills for Repeatable Work
28:11 CRM Reality Check
30:10 Greenfield vs Brownfield
32:46 Distribution Is the Moat
34:35 Wrap and Next Demos

Doom and Q discuss how fast AI tooling is changing content creation, coding, and go-to-market. They troubleshoot audio/headphone issues, then dive into experiments with Claude: generating markdown-based enablement decks, comparing Gamma to Claude for PowerPoint, creating editable master slide layouts, and quickly opening/editing outputs in Google Slides. They describe building a React/JavaScript version of a security enablement deck, experimenting with an in-browser presentation skill, and trying to set up OpenClaude with an LMS and Discord but hitting login/token problems. The conversation covers Figma/Canvas-style workflows that connect AI-generated code to design iteration, and an “11 labs barbell” GTM strategy combining product-led growth with enterprise sales while lacking a middle-market motion. They discuss rising personal spend on AI subscriptions, the surge of “personal email” signups driven by “nerds at night,” and how AI lowers barriers for learning and building while deployment, security, authentication, and cloud costs remain hard. They outline a workflow to enrich personal signups into company prospects (via LinkedIn/ZoomInfo-style enrichment and CRM mapping) to identify multiple users inside the same target account and convert usage into sales outreach. They close by arguing differentiation will come from distribution (audience/platform) and the “harness” around models—prompting/context, evals, and orchestration—illustrated by performance differences between tools using the same model, plus a note on influencer distribution with a MrBeast/Salesforce-Slack ad example and encouragement to keep tinkering and building.
 
00:00 Cold Open
00:43 AI Tooling Overload
01:50 Claude Meets Figma
04:03 PowerPoint Automation
07:28 Gamma Versus Claude
09:44 Nerds at Night
13:02 High Agency Execution
15:45 Personal Emails to Revenue
19:19 Harness and Distribution
23:37 Wrap Up and Build