The Neuron: AI Explained

OpenAI and Thrive Holdings built Tax AI, a Codex-powered agent that helps prepare complex tax returns while preserving evidence for accountant review.

In this episode, Corey and Grant talk with OpenAI’s John de Wasseige and Arthur Fernandes Araujo about how expert corrections become structured signals, how Codex turns repeated failures into evals and scoped engineering tasks, and why the best AI deployments still need humans close to the work.

They also dig into what this pattern could mean for bookkeeping, audits, IT help desks, and other expert workflows where the system can measure what “right” looks like.

Relevant links:
OpenAI Tax AI case study: https://openai.com/index/building-self-improving-tax-agents-with-codex/
OpenAI Codex: https://openai.com/codex/
Harness engineering: https://openai.com/index/harness-engineering/
Thrive Holdings: https://www.thriveholdings.com/
Crete: https://www.cretepa.com/

Subscribe to The Neuron newsletter: https://theneuron.ai

Humanoid robotics challenges go beyond movement and servo motors.

The hardest problems are often AI problems. Bringing intelligence into the physical world means dealing with gravity, friction, uncertainty, and real consequences. Mistakes can break hardware.

This week on Neuron Live, we’re joined by Nikita Rudin, Co-founder and CEO of Flexion Robotics, to unpack what it actually takes to build intelligence for humanoid systems.

What we’ll cover:
🤖 Training control policies and perception models
🧪 Bridging simulation and the real world (sim-to-real)
🛠️ Robotics training pipelines and the embodied AI stack
🔮 Where humanoid and physical AI is headed next

If large language models are the brain in the cloud, what does intelligence look like when it has to walk, grasp, and not fall over?

Expect a deep dive into embodied AI, physical AI, and the systems powering the next generation of humanoid robots.

Larry Meadows, Head of Product Strategy & Evangelism for HP's Workforce Experience Platform (WXP), joins us to break down how HP is using AI to predict and prevent IT problems before employees ever notice them.

We get a live demo of the platform—from AI-powered software recommendations across 50M+ devices to automated remediation in 3–4 clicks—and dig into the global memory crisis, shadow AI risks, and why IT leaders are drowning in portals.

Whether you manage a fleet of 50 devices or 50,000, this one's worth your time.

HP WXP: https://www.hp.com/wxp
The Neuron newsletter: https://theneuron.ai

Confused by AI Skills, Projects, Gems, Custom GPTs, and Agents?

You're not alone. It's important to separate which is best for which use-case, because each has a place depending on what you're trying to get done.

In this beginner-friendly live episode, we break down what these AI terms actually mean, who creates them, and when everyday users should use each one.

Think of this as your plain-English map to the new AI assistant world:

✅ Projects = places to organize ongoing work
✅ Gems & Custom GPTs = reusable custom assistants
✅ Skills = reusable instructions and workflows
✅ Agents = AI systems that can take actions on your behalf

By the end of this live session, you'll understand the practical difference between creating a custom assistant, organizing work in a project, giving AI a repeatable skill, and letting an agent complete tasks for you.

What You'll Learn:
🔹 What an AI Skill is
🔹 What an AI Project is
🔹 What Google Gemini Gems are
🔹 What Custom GPTs are
🔹 What AI Agents are
🔹 Which one beginners should start with
🔹 The simple framework for choosing the right tool for your workflow

Get beginner-friendly AI explainers, practical tutorials, and daily updates on what matters in AI.

📩 Subscribe to our newsletter: https://www.theneuron.ai

AI can write code, pass exams, and summarize the web, but ask it to reason through a real-world image, and the magic often breaks. Andrew Dai, co-founder and CEO of Elorian, joins The Neuron to explain why visual reasoning may be one of the biggest unsolved problems in AI.

Andrew spent years at Google Brain and DeepMind, including work connected to Gemini and sparse mixture-of-experts systems. Now, he’s building Elorian around a simple but powerful idea: if AI is going to understand the physical world, it needs more than text-based reasoning layered on top of images.

In this episode, Corey and Grant talk with Andrew about why frontier models struggle with counting, navigation, design, engineering, charts, and physical reasoning; why scaling language models hasn’t solved vision; what a “visual chain of thought” might look like; and how better visual reasoning could accelerate robotics, satellite analysis, product design, and mechanical engineering.

Sponsored by Dell Technologies and NVIDIA. Learn more at techrepublic.com/hubs/the-enterprise-guide-to-scalable-ai/.

Sponsored by Outshift: Visit https://outshift.cisco.com/?utm_campaign=fy26q3_outshift_ww_paid-media_ioc-neuronai-outshift_podcast&utm_channel=podcast&utm_source=podcast to learn more about the Internet of Cognition.

Subscribe to The Neuron for more conversations with the people building the future of AI.