AI Crosses from Exploration into Engineering

The Thoughtworks Technology Radar has always been more than a list of tools. It captures what happens when technology moves from experimentation to everyday practice. In its 33rd edition, one shift stands out: AI has crossed from exploration into engineering.

Developing with Agents

Earlier Radar editions focused on Generative AI as a productivity booster. Now, AI shows up as an architectural force. The emergence of agents, powered by the Model Context Protocol (MCP), points to a world where software components reason, negotiate, and collaborate — not just execute.

The newest radar entries illustrate this change:

• Context Engineering — Designing what an AI knows and how it interprets its world.
• Agent-to-Agent (A2A) Protocol — Allowing autonomous agents to coordinate directly.
• AG-UI Protocol — Connecting agents to interfaces and user experiences.
• AGENTS.md — A simple way to describe an agent’s purpose, scope, and interface.
• Anchoring coding agents to a reference application — Grounding agents in a real, trusted codebase.

Together, they signal a maturing ecosystem of connected AI systems — where context, not code, becomes the main design material.

AI Coding Workflows

The Radar also documents how AI is reshaping software delivery workflows. It’s no longer about asking a copilot to complete a line of code. Instead, teams are learning how to engineer entire flows that blend human insight with machine assistance.

Among the new techniques:

• Spec-driven development — Define the intent before the implementation.
• Using GenAI to understand legacy codebases — Making modernization less painful and more informed.
• GenAI for forward engineering — Translating business concepts into architecture proposals.
• Curated shared instructions for software teams — Establishing collective knowledge for AI-assisted work.
• Self-serve UI prototyping with GenAI — Letting teams explore product ideas through conversation.
• Context7 — Treating context as a first-class engineering concern.

This is AI entering the SDLC — not as an afterthought, but as part of design, discovery, and delivery.

New AI Antipatterns

As always, the Radar balances enthusiasm with realism. Several AI-related antipatterns warn of the new kinds of technical debt emerging:

• Complacency with AI-generated code — When speed overshadows quality.
• AI-accelerated Shadow IT — Local scripts and agents outside governance.
• Naive API-to-MCP conversion — Wrapping legacy APIs without redesign.
• Text-to-SQL — Giving models too much direct power over data.

Each of these reflects a familiar lesson: AI doesn’t remove the need for good engineering — it magnifies both discipline and disorder.

Conclusion

Radar 33 captures a turning point: AI is no longer a lab experiment or a productivity add-on — it’s becoming a core capability of modern software engineering. We are learning not just to use AI, but to engineer with it.

Based on Thoughtworks Technology Radar 33, November 2025.