Subvolts take AI is advancing rapidly, especially in coding, with human-level performance near. Oct 2, 2025 - Developers
- Learn about Claude Sonnet 4.5 and how AI agents maintain coherency for complex coding tasks.
- Business
- Understand the rapid pace of AI development and its potential impact on computer-facing tasks and economic output.
A researcher from a leading AI lab provides an inside look at a powerful new coding model and its implications.
Sonnet 4.5 & the AI Plateau Myth — Sholto Douglas (Anthropic)
Audio streams from Matt Turck. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Sholto Douglas of Anthropic discusses Claude Sonnet 4.5, a leading coding model. He predicts AI may match human performance on computer tasks within 2-3 years. The episode explores why reinforcement learning on language models succeeded in 2024, agent coherency, and the ongoing progress without a plateau, challenging the idea of an AI slowdown. Douglas also shares his career path and insights into AI lab culture.
An AI researcher explains why AI models are getting much better at tasks like coding, predicting they'll soon perform as well as humans. This matters because it means AI could soon help us build even better AI and drive big changes in technology.
This episode offers a researcher's perspective on rapid AI progress, particularly in coding. It's for anyone tracking AI's trajectory, especially developers and business leaders interested in near-term capabilities. The discussion on coding models and agent advancements connects directly to current work at Google Gemini, Meta AI, and OpenAI, highlighting the race for more capable AI systems. Douglas's insights into Anthropic's focus on coding as a path to economic impact and AI-assisted research are particularly relevant.
- HostMatt Turck
- GuestSholto Douglastop AI researcher, Anthropic
- Guest: Sholto Douglas (Anthropic)
- Title: Sonnet 4.5 & the AI Plateau Myth
- Show: The MAD Podcast
- Length: 70 minutes
- AI Performance Prediction: 2-3 years to match human-level on most computer-facing tasks
- Key Development: Reinforcement learning on language models started working in 2024
Sholto Douglas, a top AI researcher at Anthropic, discusses the breakthroughs behind Claude Sonnet 4.5—the world's leading coding model—and why we might be just 2-3 years from AI matching human-level performance on most computer-facing tasks.
You'll discover why RL on language models suddenly started working in 2024, how agents maintain coherency across 30-hour coding sessions through self-correction and memory systems, and why the "bitter lesson" of scale keeps proving clever priors wrong.
Sholto shares his path from top-50 world fencer to Google's Gemini team to Anthropic, explaining why great blog posts sometimes matter more than PhDs in AI research. He discusses the culture at big AI labs and why Anthropic is laser-focused on coding (it's the fastest path to both economic impact and AI-assisted AI research). Sholto also discusses how the training pipeline is still "held together by duct tape" with massive room to improve, and why every benchmark created shows continuous rapid progress with no plateau in sight.
Bold predictions: individuals will soon manage teams of AI agents working 24/7, robotics is about to experience coding-level breakthroughs, and policymakers should urgently track AI progress on real economic tasks. A clear-eyed look at where AI stands today and where it's headed in the next few years.
Anthropic
Website - https://www.anthropic.com
Twitter - https://x.com/AnthropicAI
Sholto Douglas
LinkedIn - https://www.linkedin.com/in/sholto
Twitter - https://x.com/_sholtodouglas
FIRSTMARK
Website - https://firstmark.com
Twitter - https://twitter.com/FirstMarkCap
Matt Turck (Managing Director)
LinkedIn - https://www.linkedin.com/in/turck/
Twitter - https://twitter.com/mattturck
(00:00) Intro
(01:09) The Rapid Pace of AI Releases at Anthropic
(02:49) Understanding Opus, Sonnet, and Haiku Model Tiers
(04:14) Shelto's Journey: From Australian Fencer to AI Researcher
(12:01) The Growing Pool of AI Talent
(16:16) Breaking Into AI Research Without Traditional Credentials
(18:29) What "Taste" Means in AI Research
(23:05) Moving to Google and Building Gemini's Inference Stack
(25:08) How Anthropic Differs from Other AI Labs
(31:46) Why Anthropic Is Laser-Focused on Coding
(36:40) Inside a 30-Hour Autonomous Coding Session
(38:41) Examples of What AI Can Build in 30 Hours
(43:13) The Breakthroughs That Enabled 30-Hour Runs
(46:28) What's Actually Driving the Performance Gains
(47:42) Pre-Training vs. Reinforcement Learning Explained
(52:11) Test-Time Compute and the New Scaling Paradigm
(55:55) Why RL on LLMs Finally Started Working
(59:38) Are We on Track to AGI?
(01:02:05) Why the "Plateau" Narrative Is Wrong
(01:03:41) Sonnet's Performance Across Economic Sectors
(01:05:47) Preparing for a World of 10–100x Individual Leverage
- reinforcement learning
- Sholto Douglas
- a way AI learns by trial and error, getting rewarded for good actions.
- agents
- Sholto Douglas
- AI programs that can perform tasks autonomously.
- priors
- Sholto Douglas
- pre-existing assumptions or knowledge built into an AI system.
Questions people ask
- When will AI match human-level performance on computer tasks?
- Sholto Douglas predicts AI may reach human-level performance on most computer-facing tasks within the next 2-3 years.
- Why is Anthropic focused on coding?
- Anthropic focuses on coding because it is seen as the fastest path to both significant economic impact and advancing AI-assisted AI research.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Matt Turck.