I reflected on four decades of technology cycles and noted that the current AI cycle differs from previous “AI summers” and “winters.” The focus has shifted from merely building large language models (LLMs) to creating frameworks that turn LLMs into actionable agents.
If the last eighteen months were about the engine—the breathtaking, almost vertical ascent of Large Language Models (LLMs)—the next era will be about the car we build around it.
- Early models, for all their fluency, felt like they were operating with a kind of impostor syndrome, optimized for credibility over correctness.
I decided to adopt agentic frameworks over static automation to achieve adaptability and digital common sense.
I prioritized clean data, consistent metadata, and well‑defined APIs before deploying agents to avoid “garbage‑in, plausible‑garbage‑out” failures.
- Build with Purpose – align agents with clear client problems.
In an experiment, the agent completed a charity‑website migration in approximately twenty minutes and delivered a fully deployable package with impeccable documentation and unit tests.
- Look Around Corners – anticipate failure modes and design resilience.
- Inspire Trust – maintain transparency and accountability.
- New “reasoning” models offer chain‑of‑thought capabilities that resemble nascent cognition.
- Agentic coding tools can autonomously perform complex tasks but still require human supervision.