Introduction

Hello, everyone.

My name's Andrew.

I'm founder of 60 Seconds.

About 60 Seconds and Today’s Demo

So 60 Seconds is a B2B attraction go-to-market platform where we've got a load of technology that helps make the sales process easier before and potentially after the call as well.

Now, I've been tasked with doing a demo of AI agents, workflows, things I've been working on in my free time as well as in the business, and the way that I've increased my productivity and planning to increase the productivity across the business as a whole.

So if anyone caught Mark Zuckerberg's demo of the AI metaglasses, I'm hoping to fuck up slightly less than him today.

But yeah, I've got a couple of different things that I think will be interesting if they work.

If they don't work, then it was nice to know you.

So I've got a few different slides.

But the main thing is I've worn most every single hat in the business.

So I started with next to no money, literally no money.

And then growing the business with no money, it's very difficult to hire people.

So I've had to become proficient at the job, take someone on, give them a standard process, and then take them through how to be successful in their role to generate more business, then hire and keep going.

So if say we're 10 years in, we're 10 people strong, but we've also got AI agents that are assisting us across the business and assisting our clients.

And this is the thing I think people haven't really seen, and I'm hoping to show you a snapshot of that today.

From Blog Idea to Published Post with AI Agents

Traditional workflow vs. AI-assisted workflow

So let's say if you were going to try and create a blog, you would first research.

You would then come up with the copywriter.

So that might be a different person, a different person in the team, or a subcontractor.

You'd then need to come up with the visuals.

What's it going to look like on the website?

You then need to do SEO.

That's going to be a different person with different skill sets, so that's going to come at a different budget.

You then need to publish it, send it over to the website manager.

And then you need someone in marketing to promote it.

Then you've got your final result.

Now, that could be a week's worth of work to get one blog out and promote it, or many people in one team that's going to cost you potentially hundreds of thousands of pounds a year, just to be consistent to do the bare minimum as a business to promote yourselves.

But what I want to show you today is how we don't have a marketing department, but I can pretty much hopefully

We are all marketers in our business, right?

So I hate to say, we don't have a dedicated marketing department that's spinning out blogs and the coloring in department, as many corporates would say.

So I've got a potential workflow that's hopefully going to work.

Triggering the workflow via Telegram and Google Tasks

So if we go to Telegram.

So Telegram is how I communicate with one of my agents.

So this is my agent here.

So I'm literally showing you behind the scenes.

This is my virtual assistant.

So if I execute the workflow, so this is just the test mode, I can then say,

Create a task to write a new blog.

The blog should be titled How to be more productive with AI.

Now in the blog we need to tell people how they can set up agents and where they can use something like Claude Code.

Claude Code is fantastic for giving a brief, like a job description, and then you can call upon that agent to do and perform that task pretty much autonomously.

So if you're trying to deliver an end result, AI will then look at the agents, the people,

and then be able to delegate tasks to those agents and then we want to show people how they can communicate with the agents using things like telegram using things like kanban boards using chat gpt with mcp connections it's very easy once you know how but actually knowing what to do is the difficult bit so in this blog we want to give people value on how they can do it with us with a beginner interface for someone who's never even come across ai

all the way up to someone who's pretty intermediate and wants to build full departments of agents.

This is now being transcribed, so that was a voice note.

That's now being sent to a Langchain AI agent, and that is now connected up to all my different tools here.

As you can see, it's now connected up to my Google Tasks, and it's inserted the block.

Now that blog there could be now connected to another workflow, which is monitoring the tasks, looking for keywords like blog, and that might be Claude Code.

Inside Claude Code: Agents and Slash Commands

So is anyone playing with Claude Code?

Okay, if you haven't, I'm sure you will in the future.

It's quite scary when you first get to grips because it's like the black box that you shouldn't touch on a computer.

But I'm hopefully going to show you it's very easy to use it.

So there's different things in Cloud Code which people aren't using.

One is agents.

So agents are essentially you can give someone a job description and then it will follow that like a big prompt.

So before it does anything, that job description will be pre-processed before the output.

So it's following that rule guide.

And then you've got something called slash commands.

So slash command is shortcuts.

Now, that slash command can call upon multiple agents in sequence or in parallel to completing the task that you set.

So you would say, I'm creating a blog.

Can you first get the marketing manager to talk to all the other agents, the research agent to then go out and research the content further, the copywriting agent to get the right content, and then the SEO agent, and then post it on the website?

So I've created a shortcut for today called blog check.

Now usually, after that, you can tab and you can then suggest what your prompt is.

And then again, it's pre-prompted beforehand.

So this is the bit that I'm hoping is going to work.

End-to-end blog pipeline orchestration

So it should look to Google Tasks, look for the most recent task with blog.

It's then going to extract the title, extract the description, and then delegate that to the research specialist.

Once it's done the research, it's then going to use the copywriter, it's then going to use the image finder, the SEO specialist, and I've then said, you need to wait until you've got the output from previous agent, you need to wait until you've got the output from previous agent.

So this is running pretty much in sequence.

But this is how humans in my business are interacting to complete a task and delegating it across teams.

And you can see in real time how it's doing it and doing it quickly.

So then we can let that go and that's going to run tokens.

It's going to call upon different tools.

It's going to use sequential thinking.

It's going to use things to make sure that it's understood the context and it's going to deliver correctly.

So what we're seeing is it's running through this process here.

For many years I've had to pay people to do this process over and over again, and right now I've doubled the size of my team in development terms, in marketing, to deliver consistent output.

Six months ago I couldn't write a line of code, I still can't write a line of code, but I don't have to write a line of code anymore, right?

I'm like the product manager, I know what I want at the end, so I've always worked with traditional devs to achieve

the software that we want to put out there.

But right now, faster than briefing the developers, I can come up with the first concept, the MVP.

I can show them how it should look, feel, interact and the functionality.

Applying the Same Approach to B2B Sales

Now, our core business, 60 Seconds, we're doing this but for sales.

As a B2B business, we've got to reach out to other businesses to tell them what we're doing, start a conversation.

And that process is typically cold outreach via email or cold calling.

It could be ads.

And more so now, it's multichannel.

So you can't just send an email nowadays, get through the spam filters, make an impact, and generate new business.

You've got to do five to 11 touch points, and that takes a lot of constructing together.

Data sourcing, verification, and personalization

But we broke it down into you first got to find the data, who you're reaching out to, who's your ideal customer profile, so that you can use tools, data aggregators like Apollo, ZoomInfo.

The more expensive it is, hopefully the better data.

There's a thing called intent data, doesn't really exist, but they do try and sell you it anyway.

You've then got to verify the validity of each email address because there are spam traps, there's catch-alls, there's things that will put you in blacklists and jail.

And that's not good because many of the businesses that come to us say, we're in blacklists because we've been sending thousands of emails from our core domain and we can't even get an internal email to land.

That's a big no-no and that's many of our starting clients.

We've then got AI to look at each individual profile, research them, scrape data from their website, scrape data from their LinkedIn profile, collect all the context for them reaching out to them, and then trained on our proposition and what we offer.

It's then tasked to go and have a very personal interaction with that prospect to then try and get them into what is a sales call or a webinar or somewhere where we can potentially sell to them.

Multichannel engagement and measurement

We then go and create personalized landing pages, personalized videos with personalized voiceover.

We then track their engagement.

Did they click to the page?

Did they view the video?

Did they watch 25, 50, 75, 95% of that video?

We then retarget them based on their interaction and their engagement on different channels like LinkedIn.

If I've got their email address, I can send that email address directly to LinkedIn and say, Rick has viewed my video.

He's not inquired.

Can I now send him videos on LinkedIn?

LinkedIn say, that's absolutely fine.

There's no unsubscribe, but I'm not going to tell you if Rick has interacted.

So I've got to re-win his engagement via like a lead form or a demo request.

And that's where we would hopefully get people booking a meeting.

And sales is pretty consistent.

If you do X amount of activity consistently in a good way, you'll be rewarded with X amount of leads.

Those leads will then try and be put to a proposal.

Those proposals will then close.

And if you get your numbers consistent and you do that same activity over and over again, you've got a great go-to-market strategy and that's how people grow their businesses.

Pre-call prep, in-call transcription, and post-call automation

So someone's booked in a call, but we don't end there.

We then research each prospect before the call, generate a brief for the sales rep to say who they're talking to, the pain points that they might have, and questions and discovery questions that they should address on the call.

The call then happens, and actually before the call we're doing personalised videos from the sales rep to say, hey Andrew, thank you so much for booking in a call, look forward to seeing you tomorrow, and then one hour before to make sure we've got a good show rate.

Each call we have a Fathom transcript like transcriber inside of the call and that's got all the context of the discovery.

Now most people just store their calls in Fathom but what we're doing is we're then taking out the call transcript using it for context to then hand it to what Rick was showing which was knowledge bases of custom assistants

that are trained on all our previous calls, the ones that converted, the ones that didn't convert.

And then it's got GPT-5, it's then going to analyse, like a sales coach would, like it's watched the whole call, it goes back to the sales rep and says, you did well in this, this, this, you could have asked more questions, your talk time was high, and they get a very summarised coaching for every single call they do.

Now, my sales reps are doing anywhere from five to ten calls.

I think Phil had 11 calls in yesterday.

So it's back to back half an hour slots.

Now, if I was watching all those calls, I wouldn't do anything else.

So having AI do that is key.

We don't stop there.

So we then collect all the tasks.

Did you agree to send a proposal?

Did you agree to follow up with them?

That then gets translated into their task list.

If someone said, I'll send you a proposal, it then goes into PandaDocs and generates a draft proposal and sends it to the sales rep.

All the admin either side of the call is completely automated.

CRM, MCP connections, and automated pipeline management

We've stripped back and we've built our own CRM in-house with an MCP connection.

So AI with the transcript can go into the CRM, update the pipeline, add the tasks, and keep everything completely updated, pushing them along the pipeline stages.

So it came in as a sales qualified lead.

Now it's an opportunity.

Now it's a proposal.

And then after the client signs, we can then go through streamlined onboarding and we can take the context from the onboarding that the client fills in to then coming up with the tasks to hand it off to the delivery team.

So if we pop back into the agent, it's still working through.

So we're currently at the website manager stage.

So the SEO specialist has done what it needs to do.

We've got the images, we've got the copy, we've got the research.

Publishing Stack: Strapi, APIs, and Why It Matters

The website manager is hooked up to our strappy CMS.

So we used to use Webflow.

Webflow is very much like WordPress.

We've recently rebuilt our entire website using Strapi.

The reason being is we've heard about MCPs, Strapi has an API, and the API will allow us to then control the blog posting, all the content, all autonomously, without having to redeploy the website each time.

Where This Is Going: AI Agent Departments

Where I think this is going...

is AI agent departments that assist the team, run autonomously.

They're doing all the laborious tasks that people don't like doing in the background.

And they're there for you to call upon as and when you need support.

And this is going to take a lot of time to build out for each business specific to their use case.

And that's where we're finding a lot of our time is, what is the process from start to completion?

Mapping processes and selecting models

Let's flow chart it.

Let's come up with what AI model needs to sit in each step.

When we go back to this big, long 20-stage sequence, you break it down.

Where do we find the data?

Here.

Is there an API to the email verifier?

Yes.

Connect those two together.

Scaling via automation and scheduling

And then you can have a, let's say we're scraping five, 10,000 new contacts per client.

We can have a schedule that then goes in and says, oh, we've looked at your campaign.

It needs a bit more, it needs more contacts.

Let's scrape contacts.

Let's take it to the next stage, the next stage.

And most all of this is automated.

So the reason I put that there is when we look at our business, this is what we're building out over time.

Building in-house tools, UIs, and audit logs

And we've taken all the software, all the project management tools, and we're starting to build it ourselves with our own databases, our own UIs on top.

So the UI is for the team, the database and the control of the database is for the AI agent, and together they'll work together.

We've got audit logs tracking what the team do day in, day out, so we can then use that data to then train AI to then do parts of their role that they find that they don't enjoy doing, so that we can free them up to do the enjoyable tasks.

I've given NAN to our apprentice, and in two days she managed to automate 30% of her role.

And I said, don't be scared by that, I'll take that away from you, that role will be automated, but then we'll move you into building more agents, doing the things that you enjoy.

Chief agent pattern and department heads

So each of these departments can be called upon from the chief agent.

So the chief agent will understand all the contexts.

We can converse with them using WhatsApp or Slack or Telegram.

That chief agent understands what tools it's got available, which would be the department heads.

The department heads can then be called upon.

So you could have the design leads or creative director could call into the AI head of content.

The AI head of content would have access to six sub-agents.

And if you wanted it to create a social video or you wanted it to write content, you can then say, hey, I would like the last seven days of our Google Analytics results.

I'd like to compare it to the previous month.

And it would then go through this chart and then deliver it back to wherever you've conversed with it.

Expected outcome of the demo

So I'm hoping by the end of that process, which should be about eight minutes, it would then go into our website and it would actually load on a blog, perfectly formatted for our website, and it'd be live on the website so we don't even have to deploy it.

Q&A

Whilst that's happening, time for questions.

Yes, please.

How long did it take you to build this?

The blog one?

Yeah.

Day.

Nice.

Yeah.

It's class seven of those time with the processes.

Quality gates and human-in-the-loop

How are you implementing quality gates and managing those quality gates to make sure all the outputs are really good, whether they need changes, how you ascertain where further improvements can be made on the quality outputs of the agents?

Yeah.

1So we use a human in the loop at any stage that needs any quality assurance and making sure that it goes.

So I wouldn't actually naturally post it to the website.

I mean, I would, and then Drew, who's ops, would kick my ass and say, did you even read that thing that's got on the website?

No, no, I haven't.

But it's tied to tone of voice.

I'm very good at giving the context needed to get the correct output.

So a lot of the times, we build client projects, and a lot of the times it's human in the loop before it goes into their database.

So we've just built a system that takes a Firefly's call.

After the client's finished on the call, they'll go into our custom application, choose the Firefly's call.

It will go out to a customer assistant.

It will then extract all the information with structured outputs into the forms.

They'll once over it, they'll press it, and it will go into their Airtable.

So it's always human in the loop first.

This was purely demo purposes.

But it's quite cool.

So if it does work, then great.

Scaling content and SEO automation

I can hook up a scheduler and you could do things like analyze our website for SEO and then hook you up to an MCP or an API into an SEO

tool that would then see what keywords we should go for and I can pump out thousands of potential blogs to hit like page one and the KPIs that you can give this is we want to hit page one within 90 days and it's tasked to do that and then because it's very good at SEO you can just get it to continually optimize and keep going like a 24-17.

Add-ons, MCP connectors, and sequential thinking

Are you using anything to augment cloud code?

Do you use any add-ons?

Because I've experimented quite a few myself, things like Taskmaster and things like that.

I haven't tried Taskmaster.

I do like the idea of Taskmaster.

Before Cloud Code Agents, I created my own markdown file and job specs, and I asked them all to monitor the markdown file every minute.

And the manager would delegate the tasks, and the front-end developer would then pick up the tasks, do the role, and I watched them create their own daily stand-ups.

and communicate and give each other like praise and that's what led on to the agents in terms of add-ons now we pretty much just use some mcp connectors like sequential thinking context 7 and yeah today i've just plugged in google tasks just to see if it can do on the desktop so sequential thinking there's quite there's a couple of variants there's like clear thought there's

Um, it passes, well, I'm not sure, but it passes through seven or eight different attempts of the same, um, question or, and then to find what the best output is.

And then it goes against all eight outputs and then chooses the best one.

So it takes a bit longer, but it seems to think and reason before it seems to move on to the next task.

So I saw that the image finder was doing a lot of, um, sequential thinking to make sure that it found the right keywords before it went to tools like unsplash, pexels, and pixabay to select the assets.

Keeping agents always-on

So have you got a way to keep it always on?

Because that's the frustration, like, just keep going.

I just want to talk into the top of the funnel and they just go.

So there's two ways I do that.

So one is every evening before I go to sleep, I will... Carry on.

Yeah, yeah, okay, right.

I'll start with core plan mode.

Plan mode, branches, and overnight builds

So if you do shift tab...

you've got plan mode.

You can set that up to be Opus, and then you can say, hey, I'm thinking about building this brand new feature or application.

I come up with a project requirement document.

I then copy and paste the plan out, switch it back out of plan mode to

Claude Dangerously Skip Permissions.

Has anyone played with that one yet?

Yeah, that's the only way to do it, but Drew won't let me do it.

So I even set up a shortcut called CC, and that CC is Claude Dangerously Skip Permissions.

And then I will forward slash full stack tab, and then I will paste in the plan that sometimes I read most of it.

And then...

And then I let it go.

So I let it go to a new branch.

I literally wake up to a brand new feature.

And it's getting so good because of the agents and how I've strung them together that every day I come.

So I don't come in every single day with a brand new tool.

And I'm like, hey, guys, come and see this.

Oh, for God's sakes, what have you built now?

And then I'll try and implement that in the business.

Sorry.

From mapping to building: timelines and approach

No, I was going to say, so day to build it out, which is why we procrastinate on it because I don't have that day at the moment.

But mapping out the flow first off.

that 22-step sales process with organisation, how long did it take to get that mapped out?

Two years.

You've just got to take part of your business and go, right, what needs to go next?

But if I've been in a sales role and I've gone, how do I construct that together?

First thing I did was no code, so I did something in Coda, and then it integrated an API into Coda.

I said, with this variable, a bit like an Excel, with first name, last name, email address,

go and research their website, scrape their website.

With this output, do this.

And you just break it into tiny micro pieces.

And that's the only thing that's giving us the head start is every single day I will come up with a new thing that didn't exist yesterday and add it into the business.

And that's how our business is getting better and better.

It's never the same business day on day.

It's how do I automate that thing that took me a bit of time?

Where to start if you’re non-technical

Leading on from that, so for those that are perhaps looking at your structure and all of the sequential steps that perhaps come from a non-technical background, like you said, you weren't a coder, where does someone start?

You mentioned about selling with that little part of your business, but perhaps some of the people don't run businesses.

Where would someone like that start?

Can you think of?

Yeah, really good question.

So NAN, there's a really good guy called Nate Hurt on YouTube.

And if you watch how he stitches everything together, that's a really great starting point.

NAN is, when this happens, do this.

It's a bit like Make or Zapier.

I think that's a really good place to start.

I tasked myself with building out a new CRM.

And I've probably spent the last eight months sweating over this CRM.

because it's meant to be a sales dashboard, but then it's expanded with new features and new features and new features on top of it.

Plus it's got a really great looking UI and the sales team use it all the time.

We've managed to really shorten our sales cycles.

Sales team have to do next to no admin.

Custom CRM with embedded workflows

I've even built in our own NAN slash make into the platform because it's MCP connected up to the entire database.

So if I go to workflows, I can then come up with my own workflows in here.

When a deal gets created, I want to check if the value is this high.

And now I can save templates.

And the plan is to be able to distribute these templates to new people that are going to come on.

So you'd have your own CRM that you'd be able to then say, when you get a lead, you want to welcome them.

All of these were created by AI.

So when you give it cloud code, the context of the entire code base, you can get it to be super creative and then expand on what you're thinking and then come up with something brand new.

This has got pretty much everything.

The CRM has got access to companies, contacts, deals, meetings.

It integrates with Fathom, so it pulls through all the calls.

You've got filtering.

And I've just been learning with this, and this has broken numerous times, and I've learned how to fix it.

And then you just, it's literally, I've got this problem, how do I fix it?

And the best thing is, you just ask AI, how do I fix that?

And you learn.

Final question

Is there a cybersecurity person on your team?

And that is the end of the talk.