Podcast Summary

The Twenty Minute VC (20VC) with Harry Stebbings

Episode Title: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing
Date: December 1, 2025
Guest: Jonathan Siddharth, CEO & Founder of Turing

Episode Overview

This episode centers on the evolution and future of data labeling, research acceleration, and knowledge work automation in the era of AI with Jonathan Siddharth, founder and CEO of Turing. The conversation explores the differences between data labeling and research accelerators, the validity of revenue claims in the space, the implications of automating all knowledge work, the future of SaaS in an AI-driven landscape, and what real-world AI deployment looks like. Jonathan shares insights into where value pools are shifting, what moats exist in a world where software is commoditized, and how work—and workforces—will transform as AI progresses.

Key Discussion Points & Insights

1. From Talent Marketplaces to Research Accelerators

• Defining the Shift

  • "I think the era of data labeling companies is over and it's now the era of research accelerators. All knowledge work is going to be automated. It's only a matter of time." (Jonathan, 00:00)
  • Turing is not a traditional talent marketplace—instead it’s about training superintelligence by providing research, compute, and, crucially, complex real-world data for AI models.

• Complexity of Data Has Changed

  • Early data labeling was about simple tasks (e.g., writing a basic Python function). Now, it's about multistep workflows with domain expertise (e.g., building a whole B2B healthcare app across multiple platforms).
    • "There's a shift in the data going from simple to complex...you need expert humans in every domain." (Jonathan, 05:28)

• Agents, Not Just Chatbots

  • The transition from chatbots to agentic systems requires entirely new data paradigms.
    • "We've gone from chatbots to agents…models becoming agentic, where they can execute complex multi-step workflows in a real-world business setting." (Jonathan, 05:28)

2. Reinforcement Learning (RL) Environments for Knowledge Work

• Turing creates RL environments for every workflow in every industry, allowing models to learn multistep business tasks.
  • "You'd create a mini world model with clones of these applications … with synthetic data … and you have what's called a verifier to check whether the agent completed the task." (Jonathan, 06:59)
  • This is described as being in "innings one" of acquiring the verticalized data needed for each workflow.
    • "We're still in innings one. It's going to take a while before we get all of this data into the models." (Jonathan, 11:39)

3. Turing vs. Other Providers (Scale, Surge, etc.)

• Turing differentiates itself by being an end-to-end research partner with depth in both RL environments and direct enterprise deployments—not just a data labeling shop.
  • "Turing is a fundamentally different animal...we are training superintelligence...the era of data labeling companies is over." (Jonathan, 12:18, 13:39)
• They also build custom AI models and solutions for enterprises (Disney, Pepsi, Blackrock, Johnson & Johnson), not just for the leading labs.

4. The Future of Knowledge Work and Enterprise AI Adoption

• Automation Trajectory:
  • Jonathan believes nearly all knowledge work—“any human’s job involving a screen, a keyboard, and a mouse”—will be automated within the next decade.
    • “All knowledge work is going to be automated. It's only a matter of time...that's $30 trillion of digital knowledge work.” (Jonathan, 17:33)
• Adoption Challenges for Enterprises:
  • Enterprises are slow to adopt due to internal process and data fragmentation.
    • “My hypothesis is that companies will be very slow with back office automation. But in the front office...if you can help them make more money, they’ll adopt it much faster.” (Jonathan, 19:59)
  • Shift of value from established incumbents to nimble startups is likely ("forest fire" dynamic).

5. Knowledge Transfer and Economic Shifts

• Budget Shift from Human Labor to AI:
  • In some domains (customer support, copywriting, SEO), labor budget is already moving to AI.
    • “The transfer is pretty high in areas like customer support, copywriting, SEO…in these low risk to fail areas.” (Jonathan, 22:18)
• Incremental Progress, Not Sudden Takeoff:
  • Disputes AI “bubbles:” progress will be continuous, allowing society time to adapt.
    • “I don't see an AI bubble. These models are incredibly powerful today...they're only going to keep improving.” (Jonathan, 40:00)
    • “I don't think we'll see rapid takeoff. I think we'll see incremental continuous improvement in AI...That's actually going to be great for the world.” (Jonathan, 58:23)

6. Moats and Defensibility in an AI World

• Data-Driven Feedback Loops:
  • "One moat will be data driven feedback loops…The advantage Google had was because everybody preferred Google and liked that search engine, you saw a much more representative set of queries." (Jonathan, 28:21)
• First-Mile and Last-Mile ‘Schlep’:
  • Deploying enterprise AI isn’t turn-key; it involves cleaning, structuring, and integrating data and training humans alongside models.
    • “First mile schlepp is...Our data is a mess. It's in silos, it's super fragmented... The data is all over the place." (Jonathan, 30:20)
    • "The way we do deployments is like a tandem system where you'd have a human and AI doing the same job...If the agent is right and the human is wrong, you train the human. If the human is right and the agent is wrong, you've created a data point to fine tune the next iteration of the agent." (Jonathan, 31:54)

7. Revenue, GMV, and Business Models in AI Deployment

• Revenues Are Not SaaS ARR
  • “These are not SaaS ARR numbers. These are not those types of revenues. This is a different beast… you have to consistently keep doing a good job.” (Jonathan, 33:27)
• Revenue Concentration
  • Turing and other data providers are similar to Nvidia: a few big labs generate the bulk of revenue, but the spend is so significant that this is acceptable.
    • “For Nvidia, 39% of their revenue comes from two clients…we are in the same boat as Nvidia.” (Jonathan, 37:54)

8. SaaS Will Be Transformed—or Obsolete

• SaaS Death Knell:
  • “I absolutely believe that SaaS as we know it, I think is over. I think it's completely over…many of these AI applications are incredibly easy to build on top of these LLMs.” (Jonathan, 46:36)
• Custom/Agentic Software:
  • Companies will increasingly build their own solutions, or foundation models will consume the "apps" layer.
    • "The model is all you need...If the model is agentic and integrated...you don't need anything else in the middle." (Jonathan, 47:55)
• Pushback from Harry & Guests:
  • Many businesses are not tech-savvy enough to build/maintain custom stacks; highly vertical SaaS will endure.

9. Who Wins in Data Provisioning & Next Big Opportunities

• Few Winners, Not Monopoly:
  • "I think there’ll be a few winners. For resiliency and price competitiveness, labs want several partners." (Jonathan, 56:08)
• Robotics/Embodied AI as White Space:
  • Biggest current white space is in robotics or embodied AI as a new data domain. (Jonathan, 56:42)

10. AI’s Societal Impact and The Future of Work

• Everyone Gets 100x More Productive:
  • “In a world where I'm 100x more productive, maybe I'm able to run 100 companies. Elon maybe runs 600 companies. Every human will just be so much more leveraged.” (Jonathan, 24:55)
• Chasm or Opportunity?
  • While some fear a widening gap for low-skilled populations, Jonathan is optimistic that democratizing access to “API intelligence” will empower a broader population: “For $20 a month, if you had access to the smartest experts in coding, in stem, in sales, in marketing, I feel like more people will be able to start companies and produce active, valuable work.” (Jonathan, 27:00)
• Software Engineering Will Expand—Redefined:
  • The number of software engineers will grow, but the definition will change: “...the pool of builders is going to expand way beyond people who've graduated with a four year computer science degree.” (Jonathan, 51:26)

Memorable Quotes & Moments

• On Automating Knowledge Work:

  “All knowledge work is going to be automated. It's only a matter of time.”
  — Jonathan Siddharth (00:00)

• On Why SaaS is Over:

  “I absolutely believe that SaaS as we know it, I think is over. I think it's completely over.”
  — Jonathan Siddharth (46:36)

• On AI Model “Overhang”:

  “There is a significant model capability overhang ... the full potential of the model has not been unlocked by humans yet.”
  — Jonathan Siddharth (41:40)

• On Enterprise AI Adoption Pace:

  “I believe in slow takeoff. I'm sorry to pour cold water on all the AI doomers ... but we are not in a rapid takeoff scenario.”
  — Jonathan Siddharth (11:39)

• On Moats in AI:

  “One moat will be data-driven feedback loops ... The importance of PageRank was known ... but the advantage Google had was ... data from clickstream ... that helps your algorithms improve at a much faster rate.”
  — Jonathan Siddharth (28:21)

Important Timestamps

• 00:00-05:00 — Redefining the data labeling business; intro to research accelerators.
• 06:00-11:00 — RL environments, agentic systems, domain complexity.
• 13:30-18:00 — Custom models for enterprises/verticals; FTE vs horizontal business.
• 18:30-21:30 — AI adoption: enterprise resistance vs. startup disruption.
• 22:15-24:55 — Shift of human labor budgets to AI technology; economic consequences.
• 28:00-30:20 — Defensibility via data-driven feedback loops.
• 30:20-32:10 — “First mile schlep” and deploying enterprise AI.
• 33:00-34:46 — Real revenue vs. GMV in AI businesses.
• 37:50-39:30 — Revenue concentration; government incentives for proprietary models.
• 40:00-41:40 — Dismissing the “AI bubble”; capability overhang.
• 46:36-49:20 — Why SaaS is dead; risks of agentic models replacing the apps layer.
• 51:26-52:01 — The future (and redefinition) of software engineering.
• 55:30-58:23 — Market structure; likelihood of a few big winners.
• 58:23-59:26 — Widely held AI myths: slow takeoff, incremental change.
• 63:51-65:02 — Future vision: AI amplifies human potential to superhuman levels.

Conclusion

Jonathan Siddharth paints a future where traditional data labeling and SaaS business models are relics, replaced by intelligent research accelerators and deeply integrated, agentic AI systems. The winners will be those who combine deep research capabilities, rapid adaptation, and a feedback-driven approach to both enterprise and model training. While disruption to established business processes and roles will be profound, Jonathan remains optimistic that productivity, innovation, and entrepreneurship will flourish as AI supercharges human capabilities.