What do you do when the technology you believe in prevents customers from using your product?

In this solo episode of Venture Step Podcast, I review Tae Kim's [*The Nvidia Way: Jensen Huang and the Making of a Tech Giant*](https://wwnorton.co.uk/books/9781324086710-the-nvidia-way). NVIDIA's position today can make its rise look inevitable. The book takes us back to the years when Jensen Huang, Chris Malachowsky, and Curtis Priem still had to decide whether to leave secure jobs and when early product bets nearly exhausted the company.

The NV1 and NV2 were technically ambitious, but compatibility problems made adoption difficult. I walk through the decision to invest in simulation, the pressure to redesign with little runway, and the RIVA 128 launch. For anyone building a product, the question is familiar: are you solving a problem the market can use now, or asking customers to follow you somewhere they are not ready to go?

Surviving once was not enough. I explore how a faster release rhythm, backward compatibility, and close work with PC manufacturers helped NVIDIA stay in the conversation. More releases meant more chances to learn. I connect that lesson to slower insurance product cycles and the danger of waiting years to get real market feedback.

The story also runs through relationships and people: "rough justice" in a long-term partnership with TSMC, the Xbox deal, years of persistent recruiting, Jensen's whiteboard reviews, and his "top five" updates. Then I turn to researcher feedback, CUDA, and the investment that made GPUs useful beyond graphics and central to AI work.

Listen for the painfully slow simulator, the employees who taped icons to their desks to save time, and the practical choice to cut scope while protecting a release date. My takeaway from the book: technical excellence and business judgment have to develop together. The visible victory lap rarely shows the cost of getting there.

CHAPTERS

00:00 Introduction, NVIDIA's founding, and Jensen Huang's vision

02:10 Product development and market missteps

04:56 Innovative use of emulators in chip design

07:36 Market feedback and product iteration

10:18 Staying ahead with rapid release cycles

13:39 Building long-term partnerships and distribution

16:56 Talent acquisition and company culture

20:23 The future of NVIDIA and AI innovation

23:11 Key takeaways for entrepreneurs and tech leaders

SOURCES

Tae Kim, *The Nvidia Way*:

https://wwnorton.co.uk/books/9781324086710-the-nvidia-way

NVIDIA corporate timeline, including the 1997 RIVA 128 launch:

https://www.nvidia.com/content/timeline/time_97.html

NVIDIA's CUDA overview:

https://developer.nvidia.com/hpc

What changes when you stop asking AI for answers and start handing it work?

Dalton breaks down OpenAI's DevDay announcements through a live experiment: call Dot, assign a plain-English briefing, request a sourced fact check, and ask it to design a five-second Venture Step intro. Then keep recording while the work happens and return to inspect the results.

The bigger idea is a personal chief of AI staff: one place to communicate with agents, delegate tasks, and review their output. Dalton explores why avatars could make that feel more approachable, how Space and Pages could support collaboration, and why ChatGPT's plugin ecosystem matters for founders trying to reach customers.

Enthusiasm meets friction: computer crashes, changing usage allowances, and the difficulty of keeping multiple projects organized in one chat. The intro demo shows the promise of handing off a creative task, while the follow-up conversation sharpens what was actually announced.

Two distinctions matter: OpenAI's audience claim is "our collective 1.2B weekly users," not an independently verified ChatGPT active-user count. Using a ChatGPT allowance in participating tools covers eligible AI work; it doesn't automatically pay for an entire third-party product.

Listen for "I'm recording the episode plus I'm getting work done," the live intro reveal, and the question of whether reviewing a deliverable becomes more important than steering every prompt.

CHAPTERS

00:00 DevDay and the live experiment

01:36 Calling Dot and assigning the tasks

09:51 Why AI agents have avatars

16:36 Specialized Dots and collaboration

18:09 Space and Pages

21:30 Usage allowances and developer tools

30:17 ChatGPT as a business distribution channel

34:41 Checking Dot's results

41:11 Allowance and audience-count clarifications

42:22 The intro reveal

43:11 From idea to action and the organizational gap

SOURCES

OpenAI DevDay recap, including model/API releases, Space, plugins, Sign in with ChatGPT, and Pro 500:

https://openai.com/index/devday-2026-recap/

Dots:

https://learn.chatgpt.com/docs/dots

Pages:

https://learn.chatgpt.com/docs/space/pages

Plugin review and discovery:

https://developers.openai.com/plugins/deploy/app-review

A real town square is noisy, awkward, and alive. What happens when a social feed recreates the bustle, but the people and reactions may be synthetic?

Dalton revisits his “Synthetic Rubicon” idea and asks what AI slop does to trust. He traces the term’s rise, then explains three traits that let synthetic filler spread: superficial competence, asymmetric effort, and mass producibility. Shrimp Jesus and Jean Phil show how odd content can attract attention. Michael Smith’s bot-streaming fraud and the fictional influencer Emily Hart show how attention can become money.

The episode also takes the objection seriously: people have always made mediocre content, and AI can help people learn and create. The real question is who bears the cost of checking a claim, and whether human time and in-person connection become more valuable as feeds fill up.

Moments to listen for:

• “What if that town center was social media?”

• The “whack-a-mole” gap between generating a convincing video and checking it.

• “Why am I even here?” when a feed stops showing you your friends.

Sources and examples:

• Term history: https://www.nytimes.com/2024/06/11/style/ai-search-slop.html

• Simon Willison on slop: https://simonwillison.net/2024/May/8/slop/

• Three traits of slop: https://arxiv.org/abs/2601.06060

• Shrimp Jesus research: https://cyber.fsi.stanford.edu/news/ai-spam-accounts-build-followers

• Jean Phil clip: https://www.instagram.com/reel/DdZ1-maCf_V/

• Jean Phil reporting and the token’s reported market cap: https://www.dexerto.com/tiktok/who-is-jean-phil-ai-character-behind-multimillion-dollar-meme-coin-sparks-fake-persona-trend-3412528/

• Michael Smith guilty plea: https://www.justice.gov/usao-sdny/pr/north-carolina-man-pleads-guilty-music-streaming-fraud-aided-artificial-intelligence-0

• Emily Hart investigation: https://www.wired.com/story/ai-generated-maga-girls/

• Automated web traffic (not a count of social posts): https://www.imperva.com/resources/resource-library/reports/2025-bad-bot-report/

• LinkedIn’s “AI slop” feedback option: https://www.linkedin.com/help/linkedin/answer/a15410007

• YouTube CEO’s AI-slop priorities: https://blog.youtube/inside-youtube/the-future-of-youtube-2026/

Previously on Venture Step: Episode 93, “The Synthetic Rubicon”: https://venturestep.net/episodes/the-synthetic-rubicon-crossing-the-uncanny-valley