Episode Summary
Executive Summary: The panel argues that GPT-4 with web browsing and plugins marks a major shift from static chatbots to practical work tools. Through demos in hiring, research, journalism, real estate, and video editing, they conclude AI is already automating meaningful portions of knowledge work, especially curation and routine production tasks, and will reshape team structures, productivity, and hiring decisions quickly.
Main Topics: GPT-4 web browsing as a practical work tool (Priority: 5/5): Sonny demonstrates how GPT-4 with web access can draft job descriptions, analyze code repositories, and research current news by pulling from live sources, showing its value for temporal and research-heavy tasks. AI’s impact on jobs, layoffs, and productivity (Priority: 5/5): The discussion centers on the view that AI will permanently reduce the need for many roles, especially outsourced, repetitive, and support functions, while making remaining employees materially more efficient. AI-powered media curation and journalism workflows (Priority: 4/5): Jason describes a plan to use AI at Inside to ingest past newsletters, auto-curate sources, generate first-draft summaries, and let editors focus on verification and original reporting. Plugin ecosystems and prompt engineering (Priority: 4/5): The speakers debate OpenAI plugins, their current limitations, and how prompt quality plus API design determine usefulness. They argue prompt engineering should be embedded into the product. AI video editing and content repurposing (Priority: 4/5): Opus Pro is showcased as a tool that can automatically clip long interviews into social-ready segments with subtitles and titles, potentially replacing much manual editing work. AI in consumer research and decision support (Priority: 4/5): Examples like desalination cost analysis, policing strategy prompts, and personal training prompts illustrate how users are turning ChatGPT into a rapid-answer analyst and planning assistant. Potential disruption in real estate and marketplaces (Priority: 3/5): The Zillow plugin demo shows how AI can translate natural-language housing queries into structured search, suggesting a future where brokers and listing platforms are disintermediated or forced to adapt.
Key Arguments: GPT-4 with browsing matters because it removes the model’s old cutoff-date limitation and makes it useful for current, real-world tasks. AI is not just replacing jobs directly; it is making teams smaller and forcing employees to use tools that eliminate the need for formerly outsourced or adjacent roles. The biggest near-term labor impact will likely hit low-level, repeatable knowledge work such as QA, copywriting, SDR work, and support. Productivity gains will be uneven: a minority of users and teams are already materially leveraging AI, while most have not yet adopted it. AI can absorb curation and summarization tasks, allowing human editors and operators to focus on higher-value judgment, interviews, and verification. Plugins are promising but immature; their usefulness depends on both the host model and the quality of the partner API. Future AI systems may become multi-user, collaborative workspaces with forked conversation trees and multiple AIs, not just single-user chat tools.
Data Points: GPT-4 browsing release timing: about 10 days before the episode - Sonny refers to OpenAI’s new web browsing capability as a very recent release Model cutoff date: September 2021 - The original GPT knowledge cutoff noted as a limitation Inside AI vision: 2 full-time AI developers - Jason says he plans to hire AI engineers for Inside Estimated work replaced by AI this year: 30% - Jason says he is using 30% as an internal estimate for team work replaced Possible AI replacement range per role: 10% to 50% - Jason cites Brian Chesky’s estimate for different positions Customer support automation potential: 50% - Jason attributes this estimate to Aaron Levie Current AI adoption in Jason’s company: 35% - Jason says 7 of 20 people have shown meaningful AI use Copilot X code contribution: 18% - A VP engineering at a public tech company reportedly said 18% of code comes from Copilot X OpenAI-generated clip count: 8 clips - Opus Pro turned a long interview into about eight clips Estimated clip wait time: 8 minutes - Opus queue shown during the demo Estimated clip pricing by contractors: $100–$150 per clip - Prior outsourced clip-editing offers discussed Estimated annual desalinated-water cost per U.S. person: $221.92 - Jason’s GPT-driven calculation combining per-gallon and annual usage figures Per-gallon desalination cost: $0.0076 - Used in the water cost calculation Annual water use per U.S. person: 29,200 gallons - Used in the water cost calculation LinkedIn members: 875 million - Sponsor segment cites LinkedIn user base Senior-level executives on LinkedIn: 180 million - Sponsor segment cites decision-maker reach C-level executives on LinkedIn: 10 million - Sponsor segment cites purchasing-power audience Real estate commission: 6% - Jason criticizes the standardized broker fee structure Interest in a LinkedIn ad credit: $100 - Sponsor offer for LinkedIn Marketing
Pivotal Quotes: "This is not a drill." — Jason: Opening the AI roundtable and emphasizing urgency "This is such a transcendent product that you can't shut up about it if you use it." — Jason: Discussing why ChatGPT’s usefulness itself drives growth "What he has to understand is this is such a transcendent product that you can't shut up about it if you use it." — Jason: Arguing that product quality can outweigh explicit virality features
Implications: AI is moving from novelty to infrastructure. Teams that adopt browsing, plugins, and automation will compress labor costs, speed output, and change how research, media, sales, and real estate work are done. Non-adopters risk falling behind quickly.
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.