Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

Bit of a ramble! not even the weekend!

GitHub Copilot announcement here

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

So, I guess it is inevitable, UI design is getting better and better from an AI perspective.

New Anthropic Claude Design

UX Pilot AI

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

So, WordPress has had a good run right? 24 years as of the time of this recording :)

Introducing EmDash — the spiritual successor to WordPress that solves plugin security

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

Okay, a longer (much longer) Easter 2026 version of Just Shut Up Five Mins! :)

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

Okay, most coding agents (copilot, claude code, codex etc.) have plan modes and so on, but sometimes you just need the overview perspective when creating code or adding features.

As I have been banging on for the last few weeks about SKILLS.MD (open spec for coding agents to focus your prompts and son), then these two extensions for VS Code (and therefore Antigravity, Cursor etc.) and JetBrains etc. (as well as standalone I think!) take that step a little further and makes things a little more GUI driven.

I have used Traycer before and it is great at spec driven development (as the cool kids used to say) but as of now, Verdent has some really interesting features - really like having a complete coding team.

Anyway, take a listen for more info!

.. oh and not to forget using Olama direct in VS Code (or other flavours of VSC) without extensions etc.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

Yes, skills are important - especially in the AI world and not just for coders!!

Anthropic created this open standard of skills now supported by all of the main LLM systems (or at least most and others are adding support)

For dotnet specific skills, click here

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

As we know, things are definitely moving fast in the AI world. No doubt roles are going to (if not already) change. We are moving towards becoming conductors or orchestrators of software/designs.

Google stitch is taking the design aspect one step further. Take a listen for more info!

Subscribe now

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

AI Skills - worth reading up on yourself to power your AI with tried and tested directions. Impeccable takes a frontend skill developed by Anthropic (Claude etc.) and enhances it in a very nice way.

Not sure what this is all about? take a listen and find out :)

Subscribe now

Here is the github repo

Examples and a bit more info here

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Before it was computer viruses, now it is AI manipulation, prompt injection and worse!

So, even if you are building custom GPTs, chatbots and certainly if you are a company providing AI chat interfaces, you might want to check out promptfoo.

I have seen demos of so many different platforms leaking PII (personal identifiable information) and worse, credit card details, API keys and goodness knows what!

Seems to be the new frontier to content with.

It is quite sad really that we have all of this technology, power and opportunities and yet human nature seems to want to create things to harm and destroy - yes, I am a little naive it seems!!

Anyway, take a listen - even if you are building personal AI projects… worth understanding.

Subscribe now

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

You can check out the video here showing a bit more detail

Subscribe now

Some links from the show:

Wispr (TBH, my favourite)

Typeless

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Enter a more simple approach that just uses Claude Code CLI as the brain along with some other free resources to add memory, text embeddings (meaning of words etc.) and I have a completely locked down mini virtual private server (512mb RAM, 1 vCore) running on AWS Lightsail for around $5 (free for 3 months). You can run this on your own local linux box too or even in Docker of course or anywhere a service offers linux servers (everywhere then!).

I have hooked in:

• Telegram: as my main chatbot interface which can accepts voice as well as text

• Twilio: for a real telephone number to talk/SMS with (optional, but why not!)

• Official Google MCP for calendars/email

• Official Microsoft MCP for my MS calendar/email

• I “could” easily hook into things like SalesForce, HubSpot, Slack and millions of other services.

I am currently building an Alexa skill (even though Alexa+ is coming) to give Alexa a real brain and hooks into the above.

All of this running on a very basic locked down linux VM :)

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Since recording this episode, I have seen some great examples where peeps have taken the idea of OpenClaw and created more secure (??) options using n8n or even Claude Code with Agent mode… this is something I am really interested in and will no doubt have a show out soon about that!

For now, sit back, relax, and watch those 12-13 minutes evaporate before your very eyes :)

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Skills are really useful and powerful, and in essence, just a set of folders in your code project (but not limited to code! could be a folder for marketing and what not). You can define certain aspects, detailed specifications including code/text/document snippets.

What really makes skills efficient is they a progressively loaded as needed. None of this massive context window chomping through your AI subscription tokens!

Anyway, further info below in the links:

Skills, Instructions and Moar

Anthropic Skills Info

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

This one became (unintentionally!) a bit of a weekend ramble…

Some links from the podcast:

GitHub Copilot - get all of the main AI models in one place and for $10 :)

Traycer AI - Sure, VS Code and other platforms all have skills and planning mode, but Traycer goes a wee bit further…

VS Code AI Toolkit - Seriously, the stuff you can do here and play around with is awesome!

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

However, given the modern state of Javascript (and Typescript if that is your thing), do you really need the frameworks? What about just plain web components? web components have been around for years and so has LIT (link here) to help build web components (reactive and all that good stuff). LIT is really just a small library (not a framework) weighing in at just 5kb.

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Some links from the show:

Lovable

Replit

Google Antigravity - actually, it is a downloadable app… I might have mixed up Google Firebase studio! :) which, in turn, is probably Windsurf now Google own it :)

Subscribe now

Hey, this is a free podcast. However, if you feel you want to support me, then check out Patreon. I will have some more detailed deep dives for Patreon members as well as one-to-one sessions.

Or just buy a unicorn a coffee here!

Oh, and yes, I have ended up on YouTube (doesn’t everyone eventually?):

https://www.youtube.com/@justfifteenmins but don’t worry, no ugly face to worry about (yet!).

Using Docker, you can remove a lot of the plumbing not to mention the complexity of scaling and of course security. However, my previous article about the cost implications of MCP’s is still valid - although there is light at the end of the tunnel given the article from Anthropic in my last post.

🐳 MCP via Docker Gateway

When you run MCP servers inside Docker and expose them through a Docker MCP gateway, you’re essentially creating a hub that aggregates multiple MCP servers behind one endpoint. This has some clear benefits:

• Deployment & isolation: Docker makes it easy to spin up, isolate, and manage many MCP servers without dependency conflicts.

• Scalability: You can add/remove MCP servers quickly, and the gateway abstracts away the networking.

• Consistency: Every server runs in a predictable environment, which is great for reproducibility.

⚡ Efficiency Implications

However, the token efficiency problem described in the article doesn’t magically disappear just because you’re using Docker:

• Direct tool calls (inefficient): If your agent still loads all tool definitions from the Docker MCP gateway into its context window, you’ll hit the same 150k+ token overhead. Docker just makes the servers easier to manage, but the model still sees the full set of tools downstream.

• Code execution (efficient): If you adopt the code execution approach (agents writing code to call MCP tools on demand), then Docker helps because:

  • The gateway centralizes access, so your agent can discover tools dynamically (e.g., by querying the gateway or filesystem-like structure).

  • You only load the definitions you need, reducing context usage to ~2k tokens.

  • Docker doesn’t add overhead here — it simply provides a clean way to host and scale the MCP servers.

🧩 Bottom Line

• Docker MCP gateway = orchestration convenience. It makes connecting many MCPs easier.

• Efficiency gains = only if you use code execution. If you stick with direct tool calls, Docker won’t solve the token bloat — it just aggregates the problem.

• Think of Docker as the plumbing, while code execution is the efficiency hack. Together, they’re powerful: Docker handles scale, code execution handles cost.

MCP’s (Model Context Protocol) is an ingenious invention by Anthropic (the company that created it as well as Claude, Claude code/desktop and so on) to make it much easier to connect other services to LLM’s (large language models i.e. GPT, Gemini and what not). It really does makes sense to have a common way to hook into other services be that a CRM, ERP, Calendar or whatever other service you need to connect.

However, like most technologies, things are moving fast. There are so many apps, videos and so on showing how to connect up MCP’s but there is a cost - and a rather large one!

The article (below) explains that when agents connect to many MCP (Model Context Protocol) servers, they often load all tool definitions and intermediate results directly into the model’s context window.

• This can balloon token usage to hundreds of thousands of tokens before even processing a request.

• For example, connecting to thousands of tools might require ~150,000 tokens just to load definitions.

• By shifting to code execution with MCP, agents can instead load only the tools they need and process data outside the model, reducing usage to ~2,000 tokens — a 98.7% savings in cost and latency.

The original article is here

So, still confused about all of this? fear not! here is a couple of code examples :)

These are just very trivial examples, but the idea is to show how you you can reference some MCP code “as and when” instead of pre-loading causing the excesive token cost as noted above.

Dotnet C#

Python

These examples show the code execution approach:

• Only the necessary tool definitions are loaded.

• Data flows through the execution environment instead of bloating the model’s context.

• This makes MCP-based agents far more efficient and scalable.

Who doesn’t love a chart! :)

Subscribe now