(24) Legal AI Live, February 2026, Part 2

Mar 9, 2026

Damien Riehl, a former litigator turned legal tech leader, urges precise model and version naming when judging AI. He contrasts old free chatbots with modern models like Claude Opus. The conversation covers context drift in long threads and practical fixes, plus why vibe coding and agentic browsers matter for non-technical creators.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

ADVICE

Always Name The Model When Evaluating AI

Specify which model and version when you say "AI isn't there yet".
Damien Riehl compared saying "I tried AI and it's not there yet" to confusing a 1975 car with a Ferrari, urging callers to name the exact model (e.g., ChatGPT free vs Opus 4.6).

INSIGHT

Contextual Drift Emerges Quickly In Thinking Models

Long chat threads cause contextual drift where the model's helpfulness steers it away from your intent.
Dennis Kennedy observed drift can appear after just three to five iterations, especially in "thinking" models, changing responses and ignoring recent user instructions.

ADVICE

Reset Chats Or Use Tools That Spawn New Contexts

Spawn new context windows or start fresh chats when drift appears to regain control.
Damien Riehl and others explained Opus/Claude and tools like Vincent spin up new windows or compress context to avoid accumulating drift.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

February 2026, Part 2, Top 5 Takeaways:

1. “AI is not there yet” is an Ignorant Statement.

Damien Riehl made a critical point: saying “I tried AI and it’s not there yet” without specifying which AI is like comparing a 1975 car to a Ferrari. The free version of ChatGPT from 18 months ago versus Claude Opus 4.6 today are worlds apart. Always specify which model and version you’re evaluating.

2. Context Drift is Real (But Solutions Exist).

Dennis Kennedy introduced the concept of “contextual drift”—where AI gradually strays from your original intent over long conversations, especially in thinking models. Solutions include: starting fresh conversations when you notice drift, keeping memory turned off, using tools that automatically compress and spawn new context windows (like Claude 4.6), or simply embracing shorter, focused interactions.

3. Everyone Uses AI Differently—Stop Assuming Common Experience.

A major insight: people are using completely different tools, models, workflows, and approaches. Some keep memory on, others turn it off. Some have marathon coding sessions, others use quick bursts. Anna’s comparative law research needs are vastly different from Greg’s coding workflows. There is no “standard” AI experience anymore.

4. Vibe Coding & Agentic Browsers are the Future.

Multiple panelists emphasized that voice-driven coding (vibe coding) and agentic browsers (like Comet) represent transformative capabilities. You no longer need a technical co-founder—subject matter expertise is now the key differentiator. If you can describe what you need, AI can build it.

5. Practice “Colombo-Style” Learning: Admit You Don’t Understand.

Dennis Kennedy’s closing wisdom: be like detective Colombo—be the only person willing to say “I don’t understand.” In the rapidly evolving AI landscape, intellectual humility and curiosity are more valuable than pretending to have all the answers. The willingness to not understand is what leads to breakthrough insights.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.legalailive.com