Jun 18, 2026 · 1h 10m · neon-show
Why Coding is the Fastest Path to AGI | Turing CEO Jonathan Siddharth
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The Neon Show, Turing CEO Jonathan Siddharth sits down with Siddharth Ahluwalia to discuss why coding represents the ultimate fast track to artificial general intelligence. He outlines the mechanics of LLM training, the disruption of legacy software, and the macroeconomic impact of abundant intelligence across global enterprises.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
Siddhartha as informed peer 3.7 Guest teaching 6.0 Guest disagreement 1.5 Siddhartha pushing back 1.5
speaking balance: gold is Siddhartha, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 29:28 Direct rejection of judgment obsolescence thesis
Jonathan immediately counters the host's premise that human judgment in private market investing will lose value as models grow smarter, arguing human judgment will matter more.
Hardest push from Siddhartha ▶ 1:05:00 Challenging Turing's enterprise defensibilityThe host directly contests Jonathan's business model, asking why enterprise customization won't simply be subsumed by upcoming frontier model releases.
Biggest teaching moment ▶ 11:00 Technical taxonomy of coding benchmarksJonathan educates the host and audience on precise reinforcement learning setups, covering SWE-bench, Terminal Bench, and MLE-bench Goldilocks complexity zones.
Siddhartha holds their own ▶ 53:33 Grounded VC diligence workflow breakdownThe host demonstrates domain expertise by laying out his concrete investment vetting pipeline to substantiate why pre-seed diligence requires specific data verification.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Siddhartha as informed peer | Guest teaching | Guest disagreement | Siddhartha pushing back | Why |
|---|---|---|---|---|---|---|
| Turing's Transformation: From Talent Platform to AGI Bridge | 4 | 6 | 1 | 1 | The host sets up Turing's journey from a talent platform to an enterprise AI player. Jonathan explains Turing's dual flywheel between frontier AI lab research data and enterprise deployment. | |
| Scaling High-Quality Data Engines for Coding Models | 3 | 6 | 1 | 1 | The host asks how OpenAI initially engaged Turing and what coding data entails. Jonathan articulates the role of expert vetting engines and the evolution from basic Python scripting to agentic multi-day code generation. | |
| The LLM Training Lifecycle and Coding Benchmarks | 3 | 7 | 1 | 1 | The host asks for an accessible breakdown of coding data. Jonathan breaks down the LLM training pipeline across pre-training, SFT, and RL, detailing benchmarks like SWE-bench, Terminal Bench, and MLE-bench in depth. | |
| Reinforcement Learning Paradigms and the Scaling Triad | 2 | 6 | 1 | 0 | Jonathan explains AlphaZero-style process rewards and outlines the core triad driving modern AI scaling: algorithmic research, compute, and data. | |
| Historical Inflection Points and the Power of Scaling Laws | 3 | 6 | 1 | 1 | The host asks why AI suddenly accelerated after 2022. Jonathan traces the history from ImageNet in 2012 to scaling laws and 10-trillion parameter model dynamics. | |
| Why Coding Serves as the Primary Fast Track to AGI | 4 | 6 | 1 | 1 | The host points out Anthropic's rapid ascent in coding within three years. Jonathan explains why code is the fast track to AGI due to verifiable execution and automated AI research self-improvement. | |
| Reducing Knowledge Work to Code and Computation | 4 | 6 | 1 | 1 | The host asks about Jonathan's view that all problems reduce to code. Jonathan walks through market analysis examples and lists the four core pillars of superintelligence: coding, tool use, reasoning, and multimodality. | |
| Human Agency, Judgment, and Scaling Company Output | 5 | 7 | 4 | 4 | The host suggests human judgment and agency might become obsolete if AI makes the investment decisions. Jonathan explicitly disagrees, arguing the floor rises and human judgment becomes far more leveraged, allowing individuals to run multiple companies. | |
| The 'Agent First, Human Second' Organizational Paradigm | 3 | 6 | 1 | 1 | The host asks about software engineering reductions. Jonathan explains Turing's 'agent first, human second' philosophy where humans steer while agents generate V1 across all enterprise workflows. | |
| The Dual Pincer Movement Disrupting Legacy SaaS | 4 | 6 | 1 | 1 | The host asks if the boundary between software services and products is blurring. Jonathan outlines the dual pincer movement threatening SaaS: top-down agentic capability and bottom-up custom DIY software. | |
| Enterprise Defensibility and the 'No Fine-Tuning' Camp | 4 | 6 | 2 | 3 | The host notes that GSIs rely heavily on customization and asks how new startups build systems of record. Jonathan outlines the battle between the 'no fine-tuning' frontier model camp and specialized fine-tuning. | |
| The Inference Compute Explosion and Deep Research Agents | 4 | 5 | 1 | 1 | The host asks why inference compute is surging compared to training. Jonathan uses a live VC diligence agent example to show how autonomous interview and evaluation loops consume massive inference compute. | |
| Startup Strategy and Investment Frameworks in the AGI Era | 4 | 6 | 2 | 2 | The host addresses founder anxieties regarding models consuming all software categories. Jonathan advises competing against analog markets and investing in physical or trust-based inputs and outcomes. | |
| Solving Last-Mile Enterprise Messiness with Modular AI | 5 | 6 | 3 | 5 | The host challenges Jonathan by questioning how Turing's enterprise services will stay defensible once frontier models advance into last-mile work. Jonathan explains why last-mile enterprise messiness and forward deployed engineering remain necessary. | |
| Predictions for 2035: Abundant Intelligence and Human Progress | 3 | 5 | 1 | 0 | The host asks for Jonathan's decade-out vision. Jonathan shares an optimistic prediction of abundance across drug discovery, education, and economic growth enabled by superintelligence. |
Statements from this episode (28)
Siddharth: Turing bridges frontier AI lab data and enterprise deployment
“Most data companies don't see deployment, and most deployment companies don't see data. So Turing is the one company that works with the Frontier AI Labs and Fortune 500 enterprises, and we are a trusted bridge between research and deployment.”
OpenAI partnered with Turing to teach GPT-3 coding and tool use
“OpenAI came to us when they were training GPT-III and they wanted to teach GPT-III to code. So we collaborated with them on teaching the models to code and to do tool use, function calling”
Siddharth: Frontier AI models can autonomously build software systems for weeks
“You're seeing what the coding models today are capable of building end-to-end systems, like coding for days. Sometimes a week.”
Turing crawls GitHub and buys startup code for AI training datasets
“Turing does this, where we are automatically crawling open source GitHub repos at massive scale
We are also acquiring code assets from startups and other companies, and we are running those repos and figuring out pull requests where the test cases don't pass f…”
AI reinforcement learning requires a 20% to 40% task success rate
“If you have tasks that are too easy for the model, there is no learning signal.
If it is too difficult, there is no learning signal.
So there is a sweet spot of complexity that you'd want the RL environments to be at.
Usually like, 20 to 40%, something in that…”
Siddharth: Non-binary knowledge work requires rubric-based AI evaluation
“Like with code or with math, it's relatively more binary, easy to verify. But how do you verify the quality of a board deck? Yeah. It's a, you have to be, you have to have like a good rubric based evaluator.”
Siddharth: AI scaling is constrained by compute, data, and algorithmic research
“And I think all the labs are scaling this up, and we are constrained by algorithmic research, compute, and data, and companies like Turing Advance the data pillar, NVIDIA advances the compute pillar, and of course the labs advance the algorithmic research pill…”
Siddharth: 10-trillion-parameter frontier AI models are now beginning to emerge
“The current series of frontier models are largely in the trillion parameter realm. Yeah. Now we are starting to see 10 trillion parameter models.”
Siddharth: AI scaling laws continue to hold across compute and data
“The scaling laws are continuing to hold, meaning bigger model with more data, with more compute, means that the models smoothly keep getting better. It's like the, when the pre-training loss keeps coming down, it's like the model's performance on also all thes…”
Siddharth: Verifiability makes coding ideal for reinforcement learning improvements
“Coding is one of those areas where, because it's verifiable, I think that there is a good path to using reinforcement learning to improve coding models quickly.”
AI coding improvements boost model performance across unrelated non-coding tasks
“When the models improve in coding, they have out of domain performance. They also improve in other tasks that have nothing to do with coding for reasons we don't fully understand.”
Siddharth: Turing is likely the largest coding data provider to AI labs
“We are probably the largest provider of coding data to all the labs.”
Siddharth: Coding, Tool Use, Reasoning, and Multimodality Unlock Superintelligence
“If you can solve coding, tool use, reasoning, and multimodality, you have the keys to superintelligence. Those are the building blocks. If you can solve those four, you can automate anything that a human does in front of a computer, and it'll give humans super…”
Siddharth: AI will allow individuals to run 7 to 8 companies simultaneously
“Now, I think in the limit over the next decade, every human might run seven to eight companies as these, as the loop from idea to, I mean, as you can go from prompt to company in hopefully like a single step over the next decade.”
Siddharth: AGI Will Make Humans 100x More Productive
“AGI is going to just make humans a hundred X more productive.”
Siddharth uses self-improving AI agents to help run Turing
“For example, I have a project at Turing that I run. I have a few agents running around to help me run Turing. The way I improve that code base is to also have an agent suggest to me, how should I improve the code base and to do it? So it's in a self-improvemen…”
Siddharth: Humans Should Never Create V1 of Knowledge Work
“Basically the agents create V-one of the work. Humans are verifying and iterating. Humans never create V-one.”
Siddharth: Agentic AI models will eliminate the need for intermediate SaaS
“So on one hand, the models are becoming agentic. And in theory, if the models have access to the right sources of data, You kind of don't need software in the middle. The models can do it end to end.”
Siddharth's EA built custom task-management software instead of buying SaaS
“The bottoms up pincer is every human can now write software. My EA wrote software because I have a crazy schedule. Like I work a lot. I travel a lot. Like I'm always on. So it creates a lot of work. So my EA wrote Software to manage her EA team, and that softw…”
Enterprises are abandoning fine-tuning for frontier models with context management
“What I'm seeing more and more is there's a lot more people going into the no fine tuning camp than a couple of years ago for very high value enterprise use cases.”
Siddharth built Turing's internal OS on a frontier model without fine-tuning
“I mean, I have a system at Turing that helps me run Turing, and this is a system that I built on top of one of the frontier models, and I didn't have to fine tune it. Like, all I had to do was intelligent management of context, memory, access to the right tool…”
Siddharth: Basic Workflows Suit Fine-Tuning; High-Value Use Cases Need Frontier Models
“Step one should be to, like, just do it with a frontier model and see how well it does. And most times you'll see that you don't have to touch the weights. But there are certain workflows like customer support like an invoice to pay system, some of these more …”
Siddharth: AI compute is shifting significantly toward post-training reinforcement learning
“In the past, it was a lot of the compute went into pre-training. Now a lot of compute goes into reinforcement learning in post-training as well. Especially after O-one came out and DeepSeek came out.”
Siddharth: Enterprises fail to extract full AI value due to last-mile issues
“The models are capable of so much more, but humans are still not extracting the fullest value from these models, especially in enterprise. In enterprise, I think first mile at last mile is still a problem.”
AI startups should target analog competition, not frontier AI labs
“Pick a market with weak competition. Like ideally not competition that's other big tech companies or other AI labs. Like don't pick a market with Weak competition or analog competition. For example, Uber competed with taxi companies, analog competition, not a …”
Siddharth: AI disruption of legacy SaaS applies to legacy cybersecurity products
“Legacy SaaS software also applies to legacy cybersecurity products too.”