You want to see where AI is driving us, fast? look no further than Google...
So a non-technical friend emailed me with some questions about AI and who is leading, now that Grok and Meta are dropping from the race. He asked as a question, "It seems that Anthropic is leading?"
Gmail suggested a full reply, first time. I never saw Gmail writing entire emails for me before... And Google is not holding back. Nobody is going to catch me saying either. I never saw Claude Code perform terribly, au contraire...
This is shameful. How can you trust AI writing entire documents when they can push the company's agenda at any time.
That is a big problem, a problem, if not addressed will kill that industry. Imagine if, say, a government instructs a Coding Agent (or releases a Coding Agent) that introduces backdoors in your code? Can't happen right?
Vibe coding works beautifully until it doesn't. The ceiling shows up somewhere between prototype and production: no audit trail, no review, no way to evolve the system across multiple deployments, no way to know whether yesterday's change broke today's invariants.
Spec-Driven Development is an emerging practice, but for most people it just means a collection of stale organically structured markdown documents, covering more or less the entirety of the system.
This article introduces the concept of a Sekkei (blueprint) and Generative Lifecycle Management (GLM).
Git repo for Sekkei specification and GLM Server.
Follow up article:
- Vibe Designing with Claude Code (using GLM CLI)
- How to scale GLM to a small team?
Nav has just open-sourced a new low-code project that is using GraphQL as-a-syntax to define message and event payload formats. The extensible framework currently generates json-schemas, protobufs and client libraries in golang, python and ruby for payload validation.
That problem has not really been solved yet with OpenAPI and other approaches. What is novel here is the elegance and simplicity of the GraphQL syntax used in the context of schema definitions.
Actually, TLA+ (again the way I understand it), after nearly 20 years of MDSE, convinced me that I would very rarely need to generate code. It's a much better value proposition to write correctly factored code.
TLA+ is one of the most amazing element of Computer Science. Everyone who writes code should have at least a basic understanding of it.
I have created the SAM Pattern [1] (State/Action/Model) based on my interpretation of TLA+. The goal is to make the semantics available to developers. Why not writing code as close as possible to the way it would be specified?
SAM can also be used for stateful API/Microservice orchestrations [2]
If you want a slightly more formal introduction to SAM and its relationship to TLA+ (again, based on my own interpretation) I gave this lecture last month [3].
https://github.com/jdubray