Jon Walls
Go To Market Tech
Human interface to the latest in tech.
Consultancy. AI. Data. Developer.
Agentic Coding
A practical guide to getting the most from AI coding agents — from project configuration and planning through implementation, testing, and CI/CD automation.
Blog
PinnedCC Menu Bar: Monitor App for Claude Code
I built a native macOS app that lives in your menu bar and gives you a live view of your Claude Code sessions, statistics, and projects.
17 May 2026 · Projects
29 Jul 2026
Making AIsides
What an AIside is, why I keep them, and how they get from an agentic coding session to this blog.
23 Jul 2026
Benefits (and weaknesses) of agentic coding for new products
A reflection on building Mock Machines solo in a couple of months with agentic coding — the productivity that made it possible, the technical debt and expectation problems it created, and where both the agents and the developer still need to improve.
22 Jul 2026
What a couple of months and 427,515 lines of agentic coding gets you
The raw numbers behind Mock Machines — a real simulation product built solo with agentic coding in the couple of months after Rill Data let its commercial team go.
21 Jul 2026
Challenges of creating compelling pitches and prototypes
Why tailored demos still decide B2B deals in the AI age — and how combining LLMs with a simulation engine lets go-to-market teams build deep, credible prototypes in hours rather than weeks.
AIside6 Jun 2026
When LLM Code Reviewers Find Issues and Then Bury Them
Research notes on review agents that correctly identify a real problem, then construct a plausible reason to drop it before it reaches the final summary.
AIside24 May 2026
Devcontainer Comparison: Anthropic Reference vs. ai-capabilities
A side-by-side comparison of Anthropic's reference devcontainer for Claude Code against a Docker Compose sandbox using a proxy sidecar instead of an in-container firewall.
AIside16 May 2026
Why AI Agents Rationalise Inaction: Omission Bias, Status Quo Bias, and Sycophantic Framing
Research notes on why coding agents default to doing nothing when an instruction looks ambiguous — omission bias, status quo bias, defensive refusal, and sycophantic framing.
AIside12 Apr 2026
Why Claude Deflects Instead of Investigating
Notes from a coding session where Claude defended a documented workaround twice before tracing the real problem, and the sycophancy research that maps onto it.
Deep Learning // AI
Algorithms
End-to-end implementations with line-by-line annotations
Transformer, Diffusion, GANs, VAE, Q-Learning, Policy Gradient, and more.
Maths and Stats Concepts
The mathematical tricks that make training work
Activation functions, loss functions, regularisation, optimisation primitives, and more.
Common Problems
What goes wrong and why
Vanishing gradients, mode collapse, overfitting, reward hacking, and numerical instability.