TECH BLOG
Better AI-assisted engineering starts with an engineering mindset
June 2026
How do we use AI to accelerate software delivery while maintaining engineering standards? This article explores an approach informed by Jaron Lanier’s thinking: keeping human intent, judgement and accountability at the centre. From task planning and code review to unit and integration testing, the aim is to assess AI tools by the quality and usefulness of their contribution—and verify the results.
In studying his public lectures and his foundational essay, “There Is No A.I.”, I found a fascinating argument that directly impacts how we write code: Lanier advocates that we should treat “artificial intelligence” as if it doesn’t exist. Instead of viewing an LLM as an autonomous, independent “brain”, Lanier proves that it is actually a highly advanced form of social collaboration. It doesn’t generate intelligence in a vacuum; it aggregates, compresses, and recombines the actual code and test written by millions of real humans.
When I looked at professional goals through Lanier’s lens – treating AI as an advanced multi-user version control system rather than a “magic machine” – it completely rewired how I look at automation. It turns out that the experiments I’ve been running with GitHub Copilot Agents map perfectly onto his high-level principles. I have already drafted custom skills and agents to cover these exact use cases.
Here is how Lanier’s framework maps to my desired outcome for using AI-assisted Software Development:
- Speeding Up Repetitive Tasks (with aggressive Human filtering)
- The Lanier Lens: In his video, Father of VR: They’re Creating a Future Nobody Wants | Jaron Lanier, Lanier warns against “automation bias” – the psychological trap where developers blindly accept automated code because it may look syntactically correct, lowering their own level of critical thinking.
- My Workflow Application: to avoid this trap, I have drafted GitHub Copilot “Skills” which thoroughly reviews unit tests and re-runs code coverage scripts whilst checking style, relevance, effectiveness and naming. Instead of letting AI write our unit tests, this accelerates test template generation, while maintaining strict human verification pipelines to keep quality high and errors low.
2. Testing Differences Across Models (As Data Provenance)
- The Lanier Lens: In his UC Berkeley tech lecture, Data Dignity and the Inversion of AI – Jaron Lanier, Lanier advocates for shifting our approach to using AI as a natural language prompt engineering toward “provenance calculation“. All we need to remember is different models are based on different statistical slices of human history.
- My Workflow Application: I don’t necessarily test which machine is “smarter”, but rather evaluating which model has the best training provenance for our specific code architecture. Instead of rewriting prompts to “get the right answer” we can optimise outputs by injecting local repository context (@-references), enabling the model as a targeted retrieval engine over some mythical magician.
3. Discovering Process Improvements (And Stopping Technical Debt)
- The Lanier Lens: In The Guardian, Tech guru Jaron Lanier: ‘The danger isn’t that AI destroys us. It’s that it drives us insane’ | Jar…, Lanier warns (in context of a doomsday scenario but still relatable to how we use AI), that reliance on automated systems without understanding them creates “mutually unintelligible” systems that humans can no longer manage or debug.
- My Workflow Application: When using AI for coding, it can easily spin out of control, leading to an incoherent mess that requires constant refactoring. I realised we need to use AI to better plan the code instead of just writing it. I built a first draft of a multi-step Copilot Agent workflow that takes a raw problem context, review it through the lens of a Business Analyst, analyses key components or missing requirements, and automatically create a highly structured backlog work item. I even extended it to suggest smaller manageable sub-tasks or separate work items. This is a perfect example of what Lanier calls Data Dignity – to use software to bridge human roles of Developer and BA to drive clear process improvements rather than creating opaque, unmanageable code clutter.
By adopting this “social collaboration” mindset, I can meet my goals in a way that protects long-term code quality. We should act as high-level editors of a global human knowledge base, allowing agents to help speed up our execution.
References
There is No A.I – Jaron Lanier