Thoughts For The Week - 2025.11.30
AI Coding Worfklows, Agents Elevated by MCP, Emerging AI Antipatterns
This week a deep dive into the new Thought Works Technology Radar..
AI Coding Workflows: You need to think bigger than individual coding support. How can you analyse the legacy code base and how can you setup common instructions for the whole team?
The Rise of Agents Elevated by MCP: Focus on context: What’s the best approach for you to provide the AI the needed context for the project you’re working on?
Emerging AI Antipatterns: AI supported development is still development, don’t get pulled into the standard development anitpatterns. Watch out for the return of waterfall.
1. AI coding workflows
All the early AI tools focussed on coding support for individual new green field projects. This is far from the reality of life in an existing organisation were the environment is teams (or departments) and a large existing code base. I mentioned the orchestration topic last week. And this features as a key point in the Thought Works radar…
As AI becomes strategically embedded across the software value chain — from using AI to understand legacy codebases to genAI for forward engineering — we’re learning how to better supply knowledge to coding agents.
There’s also a growing realization that AI must amplify the entire team, not just individual contributors. Techniques like curated shared instructions and custom commands are emerging to ensure equitable knowledge diffusion
Takeaway Nugget 1: You need to think bigger than individual coding support. How can you analyse the legacy code base and how can you setup common instructions for the whole team?
2. The Rise of Agents Elevated by MCP
Providing the correct context is a key driver to enabling AI to efficiently support development. MCP and agents are a proving an effective way to optimise resource usage to help support this.
We observed continued innovation in agentic workflows, where context engineering has proven critical to optimizing both behavior and resource consumption. New protocols such as A2A and AG-UI are reducing the boilerplate required to build and scale user-facing multi-agent applications. In the software development space, we compared different ways of supplying context to coding agents — from AGENTS.md files to patterns like anchoring coding agents to a reference application.
Takeaway Nugget 2: Focus on context: What’s the best approach for you to provide the AI the needed context for the project you’re working on?
3. Emerging AI Antipatterns
complacency with AI-generated code continues to be a relevant concern. Even within emerging practices such as spec-driven development, we’ve noted the risk of reverting to traditional software-engineering antipatterns — most notably, a bias toward heavy up-front specification and big-bang releases
While spec driven development can help (see last week’s post) it’s not a silver bullet. Spec-driven development is great at ensuring a consistent context for AI supported code generation, rather than random vibe coding prompts. The risk with all AI is the speed with which you can create content and therefore massive upfront specs.
I’ve lived this anti pattern this weekend, I’ve been testingAmazon’s Kiro which focusses on spec driven development. It’s far too easy to create massive spec documents. As always new requirements emerge. The actual and the planned work then diverges and you’re in exactly the “heavy up-front” risk area described by Thought Works. (At least with this test project I’m now free to start again with this in mind)
Kiro has a dedicated steering folder for project wide info and then specs are split into design, requirements and tasks which are then only used for that feature. Don’t let the individual specs get too big. Don’t revert to waterfall delivery.
Takeaway Nugget 3: AI supported development is still development, don’t get pulled into the standard development anitpatterns. Watch out for the return of waterfall!
Definitely worth reading the full article: Thought Works Technology Radar.
Have a great week.