Introduction

Setting up a live demo and goals

And we're going to try to do a live demo.

And then we're going to see how well that goes, which is probably not going to be that well.

I've got a really busy schedule coming up for the next month and a half.

And I want you to take a look at my work email and my personal email, find all of the flights that I have, and summarize them for me.

And so anyone who's seen me do a talk at this before will know that little trick that I like to pull which is basically just talking at my computer and then it kind of gives an output.

I'll talk a little bit about that for a start whilst that's kind of working.

Voice-first computing with Whisper

I basically only interact with my computer with my voice now.

I don't really type anymore.

And that's because about nine months ago, or maybe even a year ago, OpenAI released an open source model called Whisper.

What Whisper is and why it matters

And Whisper is a speech-to-text model that basically it takes your speech and it turns it into text.

And it does it incredibly fast and incredibly accurately.

And we've had speech-to-text for a long time.

It's been in like Apple, it's been in Google and so on, but it hasn't really been very good.

But this has completely changed my ability to use speech-to-text.

And as you will have seen there, it's gotten it pretty much 100% correct.

Trust, speed, and workflow changes

Within milliseconds and I am now at the point where I trust it implicitly and for especially for stuff like this because LLMs are like really capable of dealing with like oh if it missed a word or got a word wrong or whatever like I basically just talk like that and then I It puts the text in I hit enter and then I go and then I let it run and that's how I operate So I think that's pretty cool in and of itself so next

Live demo: Summarizing travel from email and calendar

So this has gone and done a bunch of work.

And so it is also searching and searching and searching.

It's looking for flights.

It's doing a bunch of stuff.

Interpreting the travel summary

And now what it's done is it's given me a summary of what my travel schedule is going to be like for the next month and a half.

Apologies in advance if this gives anyone anxiety, but I am only here for five days in the next month and a half, because my travel schedule is going to be a little bit hectic.

So I'm going to London on Monday.

Then I'm going back to London three days later.

Then I'm going back to Lisbon.

I'm going to the UK.

I'm doing a bunch of stuff.

Anyway, so basically, it summarized a bunch of stuff.

We'll not look at how much I've spent on flights.

Thank you very much.

Why integrated search and summarization is powerful

But my point here is that you could go and you could search for this in your email.

You could go and look for all of these flights and all of that kind of stuff.

But the ability for it to go and look in a system and then get all of that data and then summarize it for me is incredibly powerful and something that people don't really see too much.

Using Claude and built‑in connectors

And so I'm using Claude here.

I can't remember how much OpenAI has an interconnect, but Claude has Gmail, Google Drive, and Calendar all built in, and so you can connect it and you can be doing this stuff instantly without doing anything fancy.

It just works for you, which I think is really cool.

Weekly prep: Calendars, emails, and summaries

And so you can you can see there's some real power to that I could ask it to go and look at my calendar as well I regularly ask it at the start of a week go and look at my calendar Look at all the meetings that I have give me a summary of all those meetings then go and look in my email Find all of the emails from the people that I'm meeting this week and give me a summary of what my week is going to be so that I'm prepped for all of my meetings and

And that is kind of what I wanted to get across with some of this stuff.

It's a really powerful question.

Audience Q&A: Access and tiers

Is this feature available on the free tier?

I don't know the answer to that.

You could try it out.

Yeah, I don't know.

I'm pretty sure it's paid.

You're pretty sure it's paid?

Yeah.

So the question was, is that feature available on the free tier?

We think that it's paid.

Beyond native connectors: MCP and Zapier

So that's just like the native email and so on.

But you can also extend beyond this.

Hey, how you doing?

So you can extend beyond this.

So they've built connectors for email.

They've built connectors for a bunch of things just natively in here.

But that is Claude and Anthropic building that.

But you can also connect via MCP itself to a huge number of other systems.

And the best way that I have for people to try this out is to try it out with Zapier.

What Zapier is and how it helps

Zapier, for those that don't know, is a platform that connects

different platforms together.

So if you wanted them to connect together, but they weren't able to connect or they didn't build their own connector, that's what Zapier does.

It sits in the middle and it says when an action happens on this, do an action on this, right?

So when I get a Slack message, do an action in my email or something like that, right?

And they have 8,000 different connectors between two different systems.

Zapier’s MCP server and multi‑app workflows

And what they did, which I think is a super smart move by them, is they've built an MCP server implementation, which basically allows you to have

MCP server that you give to Claude or some other LLM, and you then add in any tool that you want from them.

So you can add in Slack, you can do all the Google Drive stuff if you want, you can add in Notion, so it allows you to connect all of these things together.

And so a good example of that is, let me see,

Maybe I won't make it run for the time being.

Example: Managing an R&D task board via MCP

But basically, here's a query that I asked it to list all the tasks in our R&D board.

So this is our backlog of tasks that we have to do in our business.

And so what I said was, can you go and list all of the tasks that we currently have outstanding in our R&D board?

And then that's what it went and did.

Here are the tasks that are currently being worked on.

Here are the tasks that are done.

Here are the tasks that are in backlog.

From lists to action: Updating and collaborating

And once I have it in that format, I can start doing really interesting things with it.

I can start saying, OK, well, for that particular task, pull all the information out.

Let's work on that a little bit.

Let's go back and forth.

can you then update that task with all of the thoughts and all of the changes that we want to make?

And you start to see that it becomes like a bit of a digital assistant, which is super powerful as well.

Case study: Automating a developer workflow

And so we do AI training for technical folks with MindStone.

And I have had boot camps that I've run where developers have used MCP.

to build a workflow which allows them to come in on Monday morning, say, go and look at the top five tasks that have been prioritized, pull those tasks down, give the AI extra context, build that context with it, and then have an AI coding agent go and write the feature or make the bug fix.

commit the code, open a pull request, do all of that whilst they're away making coffee.

And so being able to connect all of these things together is incredibly powerful and is basically what I'm trying to get through.

Case study: A ‘Find a CTO’ assistant with Notion

The final thing that I want to show you, and then we'll do questions, and then we're going to hear from Baz, is other ways of doing it, other ways of dealing with this.

So we run a fractional CTO business and we have a commercial director and he is often talking to customers and figuring out what they need from someone coming in.

A fractional CTO comes in for like two days a week and helps them like run their business.

and so what he needed was he needed to take all the requirements from a customer and he needed to figure out who are the best people for that and then build up a profile so that we could send to that customer of like here are some people that we think would be a really good fit for this so we built him this which is basically a find a CTO bot and what it does is it connects via MCP to notion where we keep

all of the availability for our CTOs as well as their skills and their experience.

And what it will do is it will go away.

It will look at that page on Notion and look at all the information in it.

It will take all of the context our commercial director has given it

it will connect those two things together and figure out, okay, John and Owen are the best people for this and then it will build up a profile for each one of them that is customised for that

customer and for what their need is.

And then we take that profile and we send that over to the customer.

And that cuts out probably an hour or two's worth of time and a conversation with me or my business partner or somebody else in the business.

And it's that interconnecting is super powerful.

Open Q&A: Context windows, connectors, and security

Any questions?

Buzz.

Yeah.

Yeah.

So Claude has been really annoying recently in that its context window is like quite short for some reason.

And so the more MCPs you use, the more your context window gets smaller.

And so that becomes a bit of a problem for sure.

And so you have to be a little bit smart about choosing which you want to, I've just turned everything on for this demo, but you get smart about turning things on and off.

I mean, this connector has like,

I don't know, 100 different things in it.

So it will start to mess with context window for sure.

Managing context limits and plugin load

The second talk I'm going to do will kind of address that, will kind of help with that.

But I don't want to give it away just yet.

So I'll come back to that in a little bit.

Toby.

Yeah.

Yeah.

Direct integrations vs. Zapier and security trade‑offs

Yeah, so it's usually better to go direct if you can, but MCP is really new, right?

And so the security around it is questionable sometimes.

And so putting it through Zapier means that you get a bit more security because they are

having to go and do some authentication with the other services and protect you a little bit.

But if you can go direct and they have set it up correctly, which probably Slack has done at this stage, then you'd be better to go direct.

Any other questions?

Sorry, say that again?

Does Make have an MCP?

Oh, I didn't know that.

That's cool.

Zapier vs. Make: Different automation philosophies

So I don't know if Make does.

So Make is different, though.

Make is an automation platform that is a little bit different to Zapier.

Like, Zapier is a little bit more event-based, and Make is to run big, long things.

So it's not really the same.

And I don't think that they have an MCP implementation or not that I've heard of anyway.

So Zapier is literally for connecting two things together.

Make is for performing operations and doing something more complicated.

Make is a little bit more like N8n rather than like Zapier.

Cool.

Conclusion and transition

All right.

Thanks.

That was talk number one.

I'll be back.

But we're going to hear from Baz now.

I'm going to run and see if the pieces have arrived and what's going on with all of those.

But yeah, thanks, everybody.