Way back in Episode One, we gazed into the crystal ball to predict exactly what the world of AI, business, and politics would look like in 2025. Now, the future has arrived. In this week’s episode, we pull up the tape, dust off those early forecasts, and ruthlessly grade our own homework.

The verdict? As Paddy puts it, "The planets have aligned." From the explosion of autonomous driving—with Waymo now hitting 450k paid rides a week—to the harsh reality of AI displacement in contact centers, many of our "bold" predictions have become standard operating procedure. We even share a real-time case study: how Claude 4.5 successfully diagnosed a failing boiler heat exchanger when a human couldn’t, proving AI is already your new local tradesman.

But we didn't get everything right. We debate the unexpected geopolitical shifts, the "absurdity" of lingering anti-AI sentiment, and the complex reality of the "LLM Wars"—where Google’s Gemini has surged to 650M users, narrowing the gap with ChatGPT.

Join us for a data-packed retrospective as we tally the wins, admit the losses, and figure out who the real winners of 2025 are (Hint: It’s NVIDIA and Starlink, definitely not Legacy Auto).

On the Scorecard this week:

• The "Prophecy" Audit: Which Episode One predictions hit the bullseye and which missed the mark?
• Real-World AI: Diagnosing heating systems via LLM & the 30% drop in customer service roles.
• Market Winners: Why NVIDIA is worth $4.7T and SpaceX is leaving NASA in the dust.
• The Political Fallout: How the rise of the far-right and Musk’s polarization affected Tesla.
• Health Tech Reality Check: GLP-1s, CRISPR, and why wearables are finally saving lives.

This week, we dive deep into the numbers behind Tesla’s autonomous driving acceleration. We look at the massive mileage milestones that are separating FSD from the competition and what the latest safety data says about the path to "unsupervised."

Then, we break down the drop of DeepSeek V3.2. This isn’t just another model release; it’s a masterclass in reinforcement learning designed specifically for agentic workloads. We unpack the benchmarks that matter, the new training methodology, and why this model might be the new standard for autonomous agents.

Finally, we turn our eyes to the grid. We discuss the UK’s energy policy following the recent budget—analyzing the mix of nuclear, gas, and renewables, and asking the hard question: Are UK consumers getting a raw deal compared to the rest of the world?

If you’ve ignored the news for the last three weeks, you’ve missed a year's worth of progress. This week, we break down the triple-threat release from Google, OpenAI, and xAI and reveal their stunning new capabilities.

In this episode:

• The New King: We analyze how Gemini 3 Pro hits record highs on the world's most challenging benchmarks.
• The Bubble Math: Hyperscalers are spending $150B+ annually on infrastructure, but is the revenue there? We discuss the "Capex Air Pocket" and whether a valuation reset is inevitable.
• The Replacement Timeline: We move past the hype to the hard numbers of displacement. We discuss the jobs most likely to be impacted, why and when.

The Takeaway: It’s no longer about who can chat the best; it’s about who can do the work.

It's all here and more on Episode 28 of the Roundtable Rethink podcast.