Six months ago, a community of developers and architects was saying: “AI slows down development by 19%.” Now those same people are building agentic flows and admitting: “knowledge of code has become worthless.” I watched this transformation from the inside - and documented it.
I’m preparing for a course launch with an education company. The founder ran a webinar series on AI in development - four sessions from September 2025 through February 2026. I watched all of them. One I attended live. A community formed around these webinars - seniors, architects, CTOs. And I watched their attitude toward AI shift from session to session.
There’s a difference between feeling “something is changing” and having 50 theses with quotes and dates. To systematize hours of spoken content from video - I had help. That’s the intrigue.
What I Did
I took transcripts from four webinars. Around 3,500 lines of text total. Ran four Claude Code subagents in parallel - each got one transcript.
Task for each subagent:
- Extract key AI-related theses with direct quotes
- Determine the speakers’ position on AI
- Document audience patterns - questions, reactions, expertise level
All four ran simultaneously. September, November, January, February - parallel, not sequential. When they finished, I compiled the results into a unified analysis: an evolution timeline, a thesis catalog, a position map.
Result
50 unique theses on AI. But the most valuable thing isn’t the count. It’s the documented transformation.
September 2025. Cautious skepticism. References to the METR study - “AI slows development by 19%.” Emphasis on risks. “Stay calm, like a bulldozer” - the mantra for the audience.
November 2025. Skepticism more detailed, but softer. Multi-agent systems, MCP, low-code entered the vocabulary. “Best practices for AI are just being born” - an acknowledgment that the topic is serious.
January 2026. Systemic realism. “AI is a platform.” “Use AI for learning.” “90% of enterprise AI initiatives failed, but 10% work.” Nuance instead of rejection.
February 2026. A roundtable with 7+ experts. And here - a shift. The speaker himself is building agentic flows. Admits: “knowledge of code and frameworks has become worthless.” Openly talks about impostor syndrome on AI topics. A group of developers and architects names Claude Code their favorite.
Six months. From “AI slows you down” to “implementation gets cheaper - what matters is WHAT you build, not HOW.” Developers and architects who six months ago cited slowdown research are now discussing token economics and constitutions for AI agents. If even skeptics with 20+ years of experience became practitioners in six months - this isn’t hype. This is a real shift.
Everyone sees that everything is changing. But so far it’s more of a feeling than a documented fact.
I needed to understand the audience before the course launch. And I ended up with a concrete example where the difference between “six months ago” and “now” is captured with quotes and dates. I know exactly what the audience has already heard. I know where they see AI’s limits. I know which topics we can build on and which are already closed.
Where You Can Apply This
In my case - partnership preparation. But the approach is broader. Analyzing a series of YouTube videos or webinars with subagents is a general way to track how someone’s position shifts over time.
Sales prep: break down a prospect’s public talks before going into a negotiation.
Competitive intelligence: how a competitor changes their narrative quarter to quarter.
Market research: take a dozen conference talks on a topic and extract trends with quotes.
The mechanic is the same everywhere: one subagent per document, in parallel, then assemble the full picture. AI systematizes, but doesn’t replace live perception - you add tone, energy, nonverbals yourself.
Sources
- How to Use Claude Code Sub-Agents for Parallel Work - Tim Dietrich
- Parallelism Meets Adaptiveness: Scalable Documents Understanding - arXiv
- Agentic Retrieval of Topics and Insights from Earnings Calls - Bloomberg/arXiv
- Building Agentic-AI Systems for Modern Competitive Intelligence - Octopus Intelligence
- Sub-Agents for Document Processing - Anthropic Claude Cookbooks
- Claude Code Sub-Agents Documentation - Anthropic
- METR Early 2025 AI-Experienced OS Dev Study