I build intelligence systems inside products and get them adopted inside enterprises.

My focus is on the stuff that determines whether AI becomes daily work or gets ignored: evaluation discipline, governance, workflow fit, rollout mechanics, and maintaining quality as things change.

Products shipped

🤖 Assist: AI operating system for marketing teams
Used by: Publicis, Toyota

🤖 BrandCheck: creative effectiveness and brand measurement
Used by: WPP, Google

🤖 PreFlight: AI video analysis and QC
Used by: WPP, Google

🤖 WAVE: AI-powered video automation
Used by: ThermoFisher

Product leader who sits close to the tech and closer to what users actually do.

Previously

🎬 Wooshii: founded to scale video operations and production
🧬 Envoke: co-founded, VR in life sciences
🍻 Hartest Brewing
🤪 Snowbombing Festival: co-founder

Patent: US 9754624 B2
Currently: CPO @ MSQ Global Studios
🏄🏽‍♂️ Surfer

If you’re turning AI into something teams trust and use daily, I’m always up for comparing notes.

Fergus Dyer-Smith

Zohe Mustafa

I Grow Your User Base w/ AI, Automation + Vibe Coding. CMO Head of Marketing / Growth Hacking / GTM. AI Generalist, AI Strategy | No-Code Workflows - Development, SaaS Founder. AI Agents, Adverts, Avatars, & Automations.

25 years in marketing.

£30M+ revenue delivered.

€65M global budgets governed.

90+ country rollouts.

I’ve led teams.
Managed agencies.
Scaled enterprise and B2B growth.

Now I build and install AI-native marketing teams.

𝗠𝘂𝗹𝘁𝗶-𝗮𝗴𝗲𝗻𝘁 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 that research,
plan, create and publish content
across every major channel.

LinkedIn.
TikTok.
Carousels.
Infographics.
Short-form video.
Long-form.

All automated.

From manual content chaos
to brand-aligned output
running daily across every channel.

No endless optimisation calls.
No dependency on freelancers.
No content bottlenecks.

Each system is trained on
enterprise-grade marketing thinking.

Built in weeks.
Configured to your brand.
Ready to run.

If you want leverage instead of headcount,
I design and deploy your AI marketing team.

𝗜𝗻𝘀𝘁𝗮𝗹𝗹𝗲𝗱. 𝗡𝗼𝘁 𝗮𝗱𝘃𝗶𝘀𝗲𝗱.

Head of Marketing,  Head of Digital Marketing, CMO, CGO, Marketing Automation

Growth Hakka is a growth hacking consultancy based in London, UK. Worked on campaigns for these industry verticals:

SAAS providers, Telecoms, Logistics, News & Media, Hospitality, Technology, FinTech, Home & Garden, Social Networks, Health and Wellbeing & E-commerce stores.

I am a no-code, growth and digital marketing, and automation specialist

I join these companies and provide marketign leadership. I can slot in as aChief Growth Officer or Chief Marketing Officer, Head of Marketing or Head of Digital Marketing manage or build their existing teams. I provide leadership, vision, plan, execution strategy, and mentoring.

My raison d'etre revolves around the principle of growth hacking. What is growth hacking? The term originated in Silicon Valley and in its essence it’s the idea that an entrepreneur or organisation can take a non-traditional approach to increasing the growth rate/adoption of their product by “hacking” something together specifically for growth purposes. Its about achieving an objective rather than following a prescribed process. Its about the intersection between Marketing, Product, Technology, Data and even HR, Finance. 

Services whilst always focused on Growth, either as a managed service or strategic input include any relevant combinations of the below:

Affiliate Marketing
A/B Testing - Product and Marketing
Analytics, Dashboards 
Conversion Rate Optimisation, 
Content Marketing
CRM
Developer Marketing
Digital: Branding, Consumer Profiling
Digital Marketing & Merchandising
Digital Strategy
Digital Media Planning & Buying across Search, Display, Mobile & Social Media
E-Commerce
Email Marketing
Gamification Strategy
Influencer Marketing
Journalist Marketing
Marketing Automation 
Mobile Marketing & Advertising
Multi-Channel Strategy
Multi Screen Campaign Strategy
Platform / Tool Selection
PR Digital Distribution
Programmatic Advertising & Marketing
SEM & SEO
Social Media Automation, Care, Enterprise, Management, Marketing & Monitoring

𝗕𝘂𝗶𝗹𝗱 𝗟𝗼𝗴 𝗗𝗮𝘆 𝟮𝟰. I’m pleased to announce my new job title.

𝗖𝗵𝗶𝗲𝗳 𝗕𝘂𝘁𝘁𝗼𝗻 𝗣𝗿𝗲𝘀𝘀𝗲𝗿.

The video is chaos
in the best possible way.

Screen one
𝗥𝗲𝗽𝗹𝗶𝘁
adding new features to JournoPitch
while I sip tea.

Screen two
𝗢𝗽𝗲𝗻𝗖𝗹𝗮𝘄
trying to take over my browser
like it pays the mortgage.
Ever so politely, mind. 

Screen three
𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲
building me a multi-agent,
multi-platform content engine
without asking for a lunch break.

And moi?

Pressing buttons.
Approving commits.
Occasionally blinking.

In a Time & Galaxy far, far away
I was writing code line by line.

Now I orchestrate.

Not replaced.
Repositioned.

Chief Button Presser.

I’ll take it.

𝗬𝗼𝘂 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗯𝗲 𝗮 𝗱𝗮𝘁𝗮 𝗮𝗻𝗮𝗹𝘆𝘀𝘁
𝘁𝗼 𝗮𝗻𝗮𝗹𝘆𝘀𝗲 𝘆𝗼𝘂𝗿 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗱𝗮𝘁𝗮.

For most companies, answers live everywhere:

  • Ad platforms.
  • Web analytics.
  • CRM.
  • Revenue data.

And analysing them together used to require:

  • Exports.
  • SQL.
  • Dashboards.

And someone technical translating it all.

𝚃̲𝚑̲𝚊̲𝚝̲ ̲𝚝̲𝚛̲𝚊̲𝚗̲𝚜̲𝚕̲𝚊̲𝚝̲𝚒̲𝚘̲𝚗̲ ̲𝚕̲𝚊̲𝚢̲𝚎̲𝚛̲ ̲𝚠̲𝚊̲𝚜̲ ̲𝚝̲𝚑̲𝚎̲ ̲𝚋̲𝚘̲𝚝̲𝚝̲𝚕̲𝚎̲𝚗̲𝚎̲𝚌̲𝚔̲.̲

Now AI agents are collapsing it.

I recently tested the AI Agent from @Coupler.io.

You can chat with your dashboard data directly inside the platform.

No SQL.
No dashboard hopping.
No switching between tools.

Just ask questions in plain language
and get answers from live data
across business apps.

Instead of reading charts,
you interrogate the stack.

No, 𝘁𝗵𝗶𝘀 𝗱𝗼𝗲𝘀𝗻’𝘁 𝗿𝗲𝗽𝗹𝗮𝗰𝗲 𝗮𝗻𝗮𝗹𝘆𝘀𝘁𝘀.
It removes friction between question and insight.

Try it for yourself: https://lnkd.in/ewJQpSkA 

What data source would you analyse first?

𝗠𝗮𝘁𝘁 𝗦𝗵𝘂𝗺𝗲𝗿 𝗷𝘂𝘀𝘁 𝘄𝗿𝗼𝘁𝗲 𝗮 𝗽𝗼𝘀𝘁 𝗰𝗼𝗺𝗽𝗮𝗿𝗶𝗻𝗴 𝗔𝗜 𝘁𝗼 𝗙𝗲𝗯𝗿𝘂𝗮𝗿𝘆 𝟮𝟬𝟮𝟬.

𝗧𝗵𝗲 “𝘁𝗵𝗶𝘀 𝘀𝗲𝗲𝗺𝘀 𝗼𝘃𝗲𝗿𝗯𝗹𝗼𝘄𝗻” 𝗽𝗵𝗮𝘀𝗲.

I read it carefully.

He’s not wrong about the speed.
He’s not wrong about capability.
He’s not wrong about the shock inside tech.
He's not wrong about the 𝗱𝗶𝗿𝗲𝗰𝘁𝗶𝗼𝗻 𝗼𝗳 𝗔𝗜 𝗯𝗲𝗶𝗻𝗴 𝗶𝗻 𝘁𝗵𝗲 𝗵𝗮𝗻𝗱𝘀 𝗼𝗳 𝘁𝗼𝗼 𝗳𝗲𝘄.

But I do diverge.

This isn’t just a capability curve. 𝗜𝘁’𝘀 𝗮 𝗴𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻.

Matt’s thesis in simple terms:

AI has crossed a threshold.
It’s compressing knowledge work.
It’s starting to build itself. Yikes!
Most white-collar roles are exposed.

That’s what the first infographic summarises.

And honestly?

If you’re not using the latest models deeply, you probably don’t realise how far this has moved.

Now my response.

Even if AI becomes better than us at most cognitive tasks…

Society cannot remove humans from the loop.

Not because we’re needed.

𝗕𝗲𝗰𝗮𝘂𝘀𝗲 𝘄𝗲’𝗿𝗲 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗱, 𝗴𝘂𝘃.

Modern systems optimise for:

• Liability
• Accountability
• Representation
• Legitimacy

𝘠𝘰𝘶 𝘤𝘢𝘯’𝘵 𝘱𝘶𝘵 “𝘵𝘩𝘦 𝘮𝘰𝘥𝘦𝘭 𝘥𝘦𝘤𝘪𝘥𝘦𝘥” 𝘪𝘯 𝘤𝘰𝘶𝘳𝘵.

You can’t elect an LLM.
You can’t fire a neural net.
You can’t hold a diffusion model responsible.

So even if AI executes better…

𝗛𝘂𝗺𝗮𝗻𝘀 𝘄𝗶𝗹𝗹 𝗿𝗲𝗺𝗮𝗶𝗻 𝗴𝗼𝘃𝗲𝗿𝗻𝗼𝗿𝘀.

This doesn’t mean jobs stay the same. It means roles mutate.

Execution compresses. Oversight expands.

From doing… to deciding.

From producing… to authorising.

From operator… to system orchestrator.

And here’s something else most doom narratives ignore.

𝗛𝘂𝗺𝗮𝗻𝘀 𝗰𝗮𝗻𝗻𝗼𝘁 𝗽𝘀𝘆𝗰𝗵𝗼𝗹𝗼𝗴𝗶𝗰𝗮𝗹𝗹𝘆 𝗰𝗼𝗽𝗲 𝘄𝗶𝘁𝗵 𝗶𝗿𝗿𝗲𝗹𝗲𝘃𝗮𝗻𝗰𝗲.

Work = identity. 
Identity = stability.
Stability = social order.

Societies don’t allow mass cognitive redundancy without redesigning the structure.

𝗪𝗲 𝘄𝗶𝗹𝗹 𝗯𝘂𝗶𝗹𝗱 𝗳𝗿𝗶𝗰𝘁𝗶𝗼𝗻. 𝗔𝘂𝗱𝗶𝘁 𝗹𝗮𝘆𝗲𝗿𝘀. 𝗠𝗮𝗻𝗱𝗮𝘁𝗼𝗿𝘆 𝘀𝗶𝗴𝗻-𝗼𝗳𝗳𝘀. 𝗛𝘂𝗺𝗮𝗻 𝗼𝘃𝗲𝗿𝘀𝗶𝗴𝗵𝘁 𝗺𝗮𝗻𝗱𝗮𝘁𝗲𝘀.

Redundant at first glance. Structurally necessary.

Now the education question.

People are asking:

“Should kids even learn to code?”

𝗠𝘆 𝗮𝗻𝘀𝘄𝗲𝗿 𝗶𝘀 𝘆𝗲𝘀.

More than ever.

AI coders like Claude Code in the hands of someone who understands code are exponentially more powerful than those who don’t.

When my son comes out of university in five years, AI will be dramatically better at coding.

But the person who understands architecture, systems, trade-offs, constraints… Will direct it.

The window right now is not about fear, we just need to learn fast.

𝗧𝗵𝗲 𝗔𝗜 𝗺𝗮𝗿𝗸𝗲𝘁 𝗱𝗶𝗱𝗻’𝘁 𝗰𝗼𝗻𝘀𝗼𝗹𝗶𝗱𝗮𝘁𝗲.
𝗜𝘁 𝗳𝗿𝗮𝗴𝗺𝗲𝗻𝘁𝗲𝗱.

I’ve just finished a 27-page report on
𝗚𝗹𝗼𝗯𝗮𝗹 𝗔𝗜 𝗗𝗲𝗺𝗼𝗴𝗿𝗮𝗽𝗵𝗶𝗰𝘀 & 𝗨𝘀𝗲𝗿 𝗣𝘀𝘆𝗰𝗵𝗼𝗴𝗿𝗮𝗽𝗵𝗶𝗰𝘀: 𝗧𝗵𝗲 𝟮𝟬𝟮𝟲 𝗟𝗮𝗻𝗱𝘀𝗰𝗮𝗽𝗲.

And one thing is clear:

We’re not in a “best model wins” era.

We’re in a demographic battle for attention.

A few interesting truths from the data:

• ChatGPT is becoming the “Google of AI”, broad, balanced, default utility

• Copilot is strongest where choice doesn’t exist,  the Fortune 500 default 

• Perplexity users spend 𝟮𝟮 𝗺𝗶𝗻𝘂𝘁𝗲𝘀 𝗽𝗲𝗿 𝘀𝗲𝘀𝘀𝗶𝗼𝗻, this isn’t search, it’s research 

• Claude Code is 90%+ technical and heavily male, the operator class 

• DeepSeek’s 18–24 cohort is reshaping the Global South engineering base 

• Character .ai users average 𝟮 𝗵𝗼𝘂𝗿𝘀 𝗽𝗲𝗿 𝗱𝗮𝘆, this is habitation, not usage 

𝘞𝘩𝘰 𝘪𝘴 𝘦𝘢𝘤𝘩 𝘱𝘭𝘢𝘵𝘧𝘰𝘳𝘮 𝘵𝘳𝘢𝘪𝘯𝘪𝘯𝘨?

  • Developers.
  • Compliance officers.
  • Students.
  • Enterprise managers.
  • Crypto-native “anti-woke” users.
  • Lonely Gen Z.

Every platform now has a psychological centre of gravity.

If you’re building product, marketing, policy, or growth,
this matters more than benchmarks.

The report breaks down:

  • Age, gender, and geo splits
  • Behavioural context and session depth
  • Psychographic identity clusters
  • The “Shadow AI” underground - this would be me if I was still 'normally' employed
  • Regional powerhouses you probably missed
  • A recommended tracking schema for 2026

27 pages. Clean breakdowns. Platform by platform.

Here is the full document.

If you work in AI, marketing, product, investing, or governance…

You’ll want to read this. Or feed to Notebook LM or other AI for a summary :-)

I've never really been bothered on “Which model is best?”

More interested in 

𝗪𝗵𝗼 𝗶𝘀 𝗲𝗮𝗰𝗵 𝗺𝗼𝗱𝗲𝗹 𝘀𝗵𝗮𝗽𝗶𝗻𝗴?

Global AI Demographics 2026 Landscape Report

𝗜 𝘀𝘁𝗮𝗿𝘁𝗲𝗱 𝗼𝗽𝘁𝗶𝗺𝗶𝘀𝗶𝗻𝗴 𝗺𝘆 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 𝗽𝗿𝗼𝗳𝗶𝗹𝗲 𝗹𝗮𝘀𝘁 𝘄𝗲𝗲𝗸.

𝗔𝘁 𝗳𝗶𝗿𝘀𝘁 𝗶𝘁 𝗳𝗲𝗹𝘁 𝘁𝗮𝗰𝘁𝗶𝗰𝗮𝗹.

Tighten the headline.
Sharpen the offer.
Remove the clutter.

25+ years of experience is a blessing.

It’s also a trap.

Because it’s easy to list everything you can do.

Harder to choose the one thing you should niche.

So I asked myself a simple question:

If the perfect inbound message landed tomorrow, what would it say?

Not:
“Can you advise us?”

Not:
“Can you run our marketing team?”

It would say:

“𝘡𝘰𝘩𝘦, 𝘤𝘢𝘯 𝘺𝘰𝘶 𝘣𝘶𝘪𝘭𝘥 𝘶𝘴 𝘢𝘯 𝘈𝘐 𝘮𝘢𝘳𝘬𝘦𝘵𝘪𝘯𝘨 𝘵𝘦𝘢𝘮?”

That was uncomfortable.
Because it meant letting go of familiar labels.

CMO.
Growth hacking.
GTM strategy.

All accurate.

But none of them were leverage.

Then I read 𝗟𝗶𝗻𝗮𝘀 𝗕𝗲𝗹𝗶𝘂𝗻𝗮𝘀’ piece on “Productise Yourself.”

𝗡𝗮𝘃𝗮𝗹’𝘀 𝗥𝗮𝘃𝗶𝗸𝗮𝗻𝘁'𝘀 framework:

𝗦𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲
+
𝗟𝗲𝘃𝗲𝗿𝗮𝗴𝗲.

And something clicked.

For many years, I’ve sold specific knowledge.

Enterprise budgets.
Global rollouts.
Agency governance.
Marketing Campaigns.
Growth Hacking strategy.
B2B revenue engines.

But it lived in my head.

Which meant it scaled with my time.

Now look at what’s happening.

Claude can code.
Agents can research.
AI can produce content at scale.
Stock Markets & Analysts are having a melt down.
The market isn’t reacting to hype.

It’s reacting to compression.

Work that required teams
now requires systems.

The value isn’t disappearing.

It’s migrating.

From labour
to leverage.

So refining my LinkedIn wasn’t about positioning.

It was about deciding which side of that shift I’m on.

𝙰̲𝚖̲ ̲𝙸̲ ̲𝚜̲𝚎̲𝚕̲𝚕̲𝚒̲𝚗̲𝚐̲ ̲𝚝̲𝚒̲𝚖̲𝚎̲?̲
̲
̲𝙾̲𝚛̲ ̲𝚊̲𝚖̲ ̲𝙸̲ ̲𝚎̲𝚗̲𝚌̲𝚘̲𝚍̲𝚒̲𝚗̲𝚐̲ ̲𝚓̲𝚞̲𝚍̲𝚐̲𝚎̲𝚖̲𝚎̲𝚗̲𝚝̲ ̲𝚒̲𝚗̲𝚝̲𝚘̲ ̲𝚒̲𝚗̲𝚏̲𝚛̲𝚊̲𝚜̲𝚝̲𝚛̲𝚞̲𝚌̲𝚝̲𝚞̲𝚛̲𝚎̲?̲

That’s why I narrowed everything to one thing:

Building AI marketing teams as systems.

Not advisory.
Not management.
Infrastructure.

Because in a world where AI can execute,
the scarce asset isn’t output.

It’s discernment.

It’s what gets built.
It’s what gets encoded.

Careers are being productised.
Companies are being compressed.

The question isn’t whether AI replaces you.

It’s whether your thinking compounds without you.

That’s the shift.
And I’m choosing leverage.

How I productised myself

𝗜 𝘄𝗮𝘀 𝗱𝗿𝗼𝘄𝗻𝗶𝗻𝗴 𝗶𝗻 𝗺𝘆 𝗼𝘄𝗻 𝗰𝗼𝗻𝘁𝗲𝗻𝘁.

𝗡𝗼𝘁 𝗯𝗲𝗰𝗮𝘂𝘀𝗲 𝗜 𝗱𝗼𝗻’𝘁 𝗸𝗻𝗼𝘄 𝗺𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴.

Because I was trying to be:

 • Strategist.
 • Writer.
 • Designer.
 • Editor.
 • Scheduler.
 • Publisher.

Across four brands.

As a team of one.

Even with 25+ years in digital and running very large budgets in the past, 

this was different.

Volume is brutal when you’re solo.

Before the agent system:

4–7 posts a week
One brand
One main channel

And that already felt like work.

Now?

40+ pieces of content
4 brands
6 channels
7 days planned in one run

 • Copy.
 • Carousels.
 • Infographics.
 • Videos. [Avatar & Faceless]
 • Scheduling.
 • Publishing.

End to end.

Did it go perfectly?

No.

The 𝚟̲𝚒̲𝚍̲𝚎̲𝚘̲ ̲𝚊̲𝚐̲𝚎̲𝚗̲𝚝̲ ̲𝚍̲𝚎̲𝚌̲𝚒̲𝚍̲𝚎̲𝚍̲ ̲𝟺̲ ̲𝚟̲𝚒̲𝚍̲𝚎̲𝚘̲𝚜̲ ̲𝚖̲𝚎̲𝚊̲𝚗̲𝚝̲ ̲𝟺̲𝟶̲.̲
𝗢𝗿𝗴𝗮𝗻𝗶𝘀𝗲𝗱 𝗺𝗮𝘆𝗵𝗲𝗺.

And what did I learn from my test run?

I can hand it over.

 • Planning.
 • Generation.
 • Distribution.

I can focus on:

Positioning.
Judgement.
Tweaks.
Direction.
Engagement.
Comments.
DMs. [I don't do ANY at the moment]

Everything else is handled.

𝗧𝗵𝗮𝘁’𝘀 𝗮 𝟱-𝟭𝟬𝘅 𝗼𝘂𝘁𝗽𝘂𝘁 𝗶𝗻𝗰𝗿𝗲𝗮𝘀𝗲 𝗳𝗼𝗿 𝗮 𝗼𝗻𝗲-𝗽𝗲𝗿𝘀𝗼𝗻 𝗼𝗿𝗴𝗮𝗻𝗶𝘀𝗮𝘁𝗶𝗼𝗻.

Same brain.
Same time.
Different leverage.

Now imagine:

A 5-person team.
A 20-person team.
An enterprise content function.

This isn’t about “using AI for posts”.

𝘐𝘵’𝘴 𝘢𝘣𝘰𝘶𝘵 𝘢𝘳𝘤𝘩𝘪𝘵𝘦𝘤𝘵𝘪𝘯𝘨 𝘢 𝘴𝘺𝘴𝘵𝘦𝘮 𝘵𝘩𝘢𝘵 𝘤𝘰𝘮𝘱𝘰𝘶𝘯𝘥𝘴.

If you’re still hesitating to give AI more control, ask yourself:

Are you protecting quality?

Or protecting your comfort?

𝗜’𝘃𝗲 𝗯𝗲𝗲𝗻 𝗵𝗼𝗹𝗱𝗶𝗻𝗴 𝗺𝘆𝘀𝗲𝗹𝗳 𝗯𝗮𝗰𝗸.

𝗕𝗲𝗰𝗮𝘂𝘀𝗲 𝗼𝗳 𝗮 𝘀𝘁𝗶𝗴𝗺𝗮 𝘁𝗵𝗮𝘁 𝗽𝗿𝗼𝗯𝗮𝗯𝗹𝘆 𝗱𝗼𝗲𝘀𝗻’𝘁 𝗲𝘃𝗲𝗻 𝗲𝘅𝗶𝘀𝘁.

I’ve been under-using AI on purpose.

Not because I don’t believe in it.

Not because it doesn’t work.

But because somewhere in the back of my mind
“𝘵𝘰𝘰 𝘮𝘶𝘤𝘩 𝘈𝘐” felt wrong.

As if using it fully meant I wasn’t really creating.

Which is ironic.

Because I already use AI every day.

I take messy notes. Half-formed ideas. Voice memos.

AI helps me structure them. Sharpen them.
𝗣𝗿𝗼𝗳𝗲𝘀𝘀𝗶𝗼𝗻𝗮𝗹𝗶𝘀𝗲 𝘁𝗵𝗲𝗺.

That’s 𝚙̲𝚛̲𝚘̲𝚍̲𝚞̲𝚌̲𝚝̲𝚒̲𝚘̲𝚗̲.

Not 𝚊̲𝚋̲𝚍̲𝚒̲𝚌̲𝚊̲𝚝̲𝚒̲𝚘̲𝚗̲.

Yet I’ve been plodding along.

One post a day. For one brand.

When I’m trying to grow 𝗳𝗼𝘂𝗿.

I don't have a human team.
No agency support.
No spare £10k per month for content ops.

And I preach leverage.

The ROI made no sense.

So, a few days ago I stopped pretending 𝗿𝗲𝘀𝘁𝗿𝗮𝗶𝗻𝘁 𝘄𝗮𝘀 𝗻𝗼𝗯𝗹𝗲.

I built a multi-brand, multi-agent content engine.

Planning agent.
Generator agent.
Scheduler agent.

First test run?

Within minutes it planned 45 pieces of content
Across 4 brands
For 7 days.

Carousels.
Video.
Infographics.
Blog posts.

All mapped to a calendar.

The generator is ready.

The scheduler is waiting.

AI isn’t creating my thinking.

𝗜𝘁’𝘀 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝗮𝗹𝗶𝘀𝗶𝗻𝗴 𝗶𝘁.

I define belief.
Positioning.
Strategy.
Narrative.

The system handles scale.

If content is your growth engine,
you don’t win by handicapping yourself.

So yes.

If you can’t beat them…

𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁 𝘁𝗵𝗲𝗺.

73% of the "must-have" AI tools from early 2025 are already gone.

Not pivoted.
Not acquired.
Gone.

And most of us never noticed.

That's not a bug in the AI ecosystem.
It might be the most important signal in it.

Every graveyard tells a story.
This one tells you which tools to trust next.

Swipe through the wreckage.
There's a pattern in there worth your attention.

No alternative text description for this image

𝗠𝗼𝘀𝘁 𝗖𝗥𝗠𝘀 𝗮𝗿𝗲 𝗻𝗼𝘁 𝗮 𝘁𝗼𝗼𝗹.

𝗧𝗵𝗲𝘆 𝗮𝗿𝗲 𝗮 𝗴𝗿𝗮𝘃𝗲𝘆𝗮𝗿𝗱 𝗳𝗼𝗿 𝗴𝗼𝗼𝗱 𝗶𝗻𝘁𝗲𝗻𝘁𝗶𝗼𝗻𝘀.

You know the feeling.

You bought the licence (s).

You sat through the onboarding. 
Long, boring, tedious

You promised the team this time would be different.

Six months later, 𝗵𝗮𝗹𝗳 𝘆𝗼𝘂𝗿 𝗰𝗼𝗻𝘁𝗮𝗰𝘁𝘀 𝗮𝗿𝗲 𝘀𝘁𝗮𝗹𝗲.

Deal stages haven't moved in weeks.

Nobody is logging calls.

And the dashboard that was supposed to give you clarity

is just a very expensive spreadsheet

with worse UX.



This is not a discipline problem.

It is a design problem.



CRMs were built for enterprise sales teams
with dedicated RevOps functions,
six-figure implementation budgets,
and the patience of a saint.


They were not built for founders.
They were not built for lean teams.
And they were absolutely not built
for the way most people actually work in 2026.

You know the secret nobody in the CRM industry
wants to say out loud:


𝗠𝗼𝘀𝘁 𝗦𝗠𝗕𝘀 𝘂𝘀𝗲 𝗹𝗲𝘀𝘀 𝘁𝗵𝗮𝗻 𝟮𝟬% 𝗼𝗳 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗽𝗮𝘆 𝗳𝗼𝗿.


The rest is 𝚜̲𝚑̲𝚎̲𝚕̲𝚏̲ ̲𝚏̲𝚞̲𝚛̲𝚗̲𝚒̲𝚝̲𝚞̲𝚛̲𝚎̲.

Expensive, underused, quietly resented shelf furniture.

I know this because I lived it.

After years across Nokia, Virgin, QVC, Accenture , eClerx
managing large budgets, corporate clients and CRM stacks
with entire teams dedicated to keeping the data clean , 

I thought I understood CRM.

Then I ran Growth Hakka.

A lean consultancy.

Real clients.

Real pipeline.

And suddenly, the tools I had relied on
inside massive organisations
felt like driving a lorry to pop to the shops.

HubSpot? Brilliant. Also overkill.

That moment crystallised something I had been

circling for years.

The problem is not that people are bad at CRM.

The problem is that CRM, as a category,

was never truly designed around how humans

want to interact with their data.


We do not think in dropdowns.
We do not speak in pipeline stages.
We think in conversations.
We think out loud.
We speak in fragments and then refine.

So that is what I started building.

AICRM is not another dashboard.

It is not HubSpot with a different colour scheme.

It is an AI-native CRM

built around how founders and small teams

actually operate day to day.

Voice-enabled.
Automation-first.
Simplified by design.

Trained to reduce friction, not add it.

You speak.
It captures.
It structures.
It follows up.

No manual data entry. Well you can if you want to
No abandoned pipelines.
No guilt-inducing dashboards
reminding you of everything you have not done.

The carousel below walks through
exactly why traditional CRM fails,
what the architecture of an AI-native alternative looks like,
and what I am building, in public, to replace it.


This is not a product pitch.

𝗜𝘁 𝗶𝘀 𝗮 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝗮𝗿𝗴𝘂𝗺𝗲𝗻𝘁.

Most AI marketing tools fail before you hit six figures in spend.

Not because they're bad tools.
Because they were never built for scale.

Here's what that actually looks like in practice.

Epic Heroes turned over £300k a year in its early years.
Built in the early 2000s, before "growth" was even a discipline.
Every system, every process, every customer touchpoint
was wired together manually.

Then I stepped into enterprise.
Nokia. Virgin. QVC. Accenture.
Budgets that ran into the tens of millions.
90+ country rollouts.
And the tools that worked brilliantly at small scale
started cracking under the pressure.

The dashboards lied.
The automations looped.
The "AI" features became liability at volume.

That pattern has never stopped repeating.

Fast forward to now.
Every week a new AI marketing tool drops.
Every founder, every CMO, every growth lead
is being sold the same promise.

"Plug this in. Watch it work."

And it does work.
For a while.
At a certain size.
With a certain level of complexity.

Then the cracks appear.

The tool can't handle multi-brand governance.
The outputs drift off-tone at volume.
The workflow that felt elegant with 10 assets
becomes a liability with 10,000.

This is not a tooling problem.
It is a systems thinking problem.

Most AI marketing tools are built for demos,
not for deployment at scale.

They optimise for the first impression,
not the fifth month of operation.

And the marketers buying them
are evaluating on features,
not on failure modes.

That's the trap.

After 25 years running marketing at scale,
from scrappy e-commerce to global enterprise,
building my own products and installing AI systems for clients,
I've developed a fairly reliable instinct
for where tools break down.

It usually happens at three points.

When the volume multiplies.
When the brand complexity increases.
When the humans step back and trust the automation.

The slide deck inside this carousel
breaks down exactly why AI marketing tools fail the scale test,
what the warning signs look like before they break,
and how to evaluate tools the way an operator would,
not the way a vendor wants you to.

This isn't theory.
It's pattern recognition earned across
£40M Nokia budgets,
£15M Virgin campaigns,
and the very real, very expensive lessons
of watching promising tech collapse
when the pressure goes up.

Swipe through.
Build smarter.

𝗜 𝗿𝗲𝗮𝗹𝗶𝘀𝗲𝗱 𝘀𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝘂𝗻𝗰𝗼𝗺𝗳𝗼𝗿𝘁𝗮𝗯𝗹𝗲 𝘁𝗵𝗶𝘀 𝘄𝗲𝗲𝗸.

𝗜’𝗺 𝘀𝘂𝗯𝘀𝗰𝗿𝗶𝗯𝗲𝗱 𝘁𝗼 ~𝟮𝟬 𝗽𝗲𝗼𝗽𝗹𝗲 𝗼𝗻 𝗦𝘂𝗯𝘀𝘁𝗮𝗰𝗸.

Paid and unpaid.

Only a handful genuinely shape how I think about AI.

One of them is Sabrina Ramonov. She shares a lot on Substack. TBH sometimes hard to keep up.

The rest?

Good.
Interesting.
Occasionally brilliant.

But collectively… overwhelming. My notis have been too much lately.
Also makes me wonder how do these folks produce sooooo much content sooooo fast!

At the same time, I’m also subscribed to a few AI writers on Medium.

I wondered do I need to get rid of one to reduce the notifications. 

And here’s the difference I finally clocked.

Substack doesn’t just deliver content.
𝘐𝘵 𝘢𝘱𝘱𝘭𝘪𝘦𝘴 𝘣𝘦𝘭𝘪𝘦𝘧 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦.

Every notification is a claim on your attention.
Every post wants to move how you think.

Medium doesn’t do that.

𝘔𝘦𝘥𝘪𝘶𝘮 𝘪𝘴 𝘢𝘮𝘣𝘪𝘦𝘯𝘵. 𝘖𝘱𝘵𝘪𝘰𝘯𝘢𝘭. 𝘚𝘬𝘪𝘮𝘮𝘢𝘣𝘭𝘦.

T'is not a quality judgement.
More a cognitive one.

Substack is where I go when I’m ready to let someone influence my mental models.

Medium is where I go to see what ideas are circulating, mutating, colliding.

The mistake I was making wasn’t subscribing to too much AI content.

It was subscribing without a system.

Now I think about it like this:

𝗠𝗲𝗱𝗶𝘂𝗺 𝗵𝗲𝗹𝗽𝘀 𝗺𝗲 𝘀𝗲𝗲 𝘄𝗵𝗮𝘁’𝘀 𝗯𝗲𝗶𝗻𝗴 𝘀𝗮𝗶𝗱 𝗮𝗯𝗼𝘂𝘁 𝗔𝗜.
𝗦𝘂𝗯𝘀𝘁𝗮𝗰𝗸 𝘀𝗵𝗮𝗽𝗲𝘀 𝘄𝗵𝗮𝘁 𝗜 𝗯𝗲𝗹𝗶𝗲𝘃𝗲 𝗮𝗯𝗼𝘂𝘁 𝗔𝗜.

Once you see that, the decision isn’t about features or price or notifications.

Medium vs Substack. Which one should claim your attention and notifications cost?

450,000 emails sent to 1,000 people in a few hours.

Not a typo.

That actually happened.
And it happened to me.

It was a HubSpot nurture campaign.
Set up with the best intentions.
Carefully crafted sequences.
Personalised touchpoints.
The full works.

But somewhere in the workflow,
a re-enrolment trigger misfired.

Each contact kept getting pulled back
into the start of the sequence.
Over and over.
Like a broken record that no one
could reach across the room to stop.

By the time we caught it,
450 emails had landed in each inbox.

High-value clients.
Senior contacts.
People who mattered.

Not exactly the "stay top of mind" strategy
we had mapped out on the whiteboard.

Here is what nobody tells you about
marketing automation:

The more powerful the tool,
the more catastrophic the failure mode.

HubSpot is not the villain here.
The workflow logic was.

We had built complexity on top of complexity.
Enrolment triggers layered over re-enrolment rules.
Suppression lists not properly cross-referenced.
A testing environment that didn't mirror production.

Classic.

And I say that without pride.

I had come from Nokia.
From Virgin.
From managing £40M budgets
across 90+ countries.

I understood systems.
Or so I thought.

The truth is, automation systems
have their own failure physics.

They will do exactly what you told them to do.
Even when what you told them
made no earthly sense at scale.

This is the part that marketers consistently miss.

We are trained to think about the message.
The subject line.
The open rate.
The CTA.

We are almost never trained to think about
the edge cases.

What happens if a contact re-qualifies?
What happens if a workflow loops?
What happens when three rules
interact in a way no one anticipated?

The answer, apparently,
is 450,000 emails.

After the panic subsided,
and after a rather uncomfortable
round of apologies to contacts
who now thought we had lost our minds,
we did what any self-respecting
systems thinker should do.

We documented it properly.
We rebuilt the architecture.
We added logic checkpoints
before every enrolment trigger.
We stress-tested against edge cases
instead of happy paths only.

And we never made that specific mistake again.

But here is the lesson worth keeping:

Automation does not make you efficient.
It makes your decisions faster.

If the decision logic is flawed,
you are not saving time.
You are scaling the error.

Every workflow you build
is a set of instructions
written to a system that has
no common sense whatsoever.

It will not pause and think
"this seems like rather a lot of emails."

It will just send them.
All 450,000.

Before you switch on the next campaign,
ask yourself one question:

If this workflow ran at ten times the volume,
with ten times the trigger frequency,
what is the worst possible outcome?

If you cannot answer that clearly,
the workflow is not ready.

Thems the breaks.
But they don't have to be yours.

𝗕𝘂𝗶𝗹𝗱 𝗟𝗼𝗴 𝗗𝗮𝘆 23 When I submitted JournoPitch to the Replit Gallery, they asked for one thing:
A screenshot proving there were no vulnerabilities.

No essay. No security questionnaire.

Just proof.

So I opened 𝘙𝘦𝘱𝘭𝘪𝘵’𝘴 𝘚𝘦𝘤𝘶𝘳𝘪𝘵𝘺 𝘚𝘤𝘢𝘯𝘯𝘦𝘳, ran it, and attached the screenshot.

𝗧𝗵𝗮𝘁 𝘄𝗮𝘀 𝗶𝘁.

What surprised me wasn’t the request.
It was how unremarkable the process felt.

Security didn’t interrupt building.
It sat quietly alongside it.

Here’s what’s doing the heavy lifting behind the scenes:

• Security Scanner
Flags vulnerable packages and dependency issues automatically
No manual scanning, no guesswork

• Secrets
API keys and credentials never live in code
You don’t “remember” to be secure. The platform enforces it

• Replit Auth
Authentication handled as infrastructure, not custom glue code
Less surface area. Fewer mistakes

This matters because the old argument is fading fast:

“Vibe-coded apps are insecure.”

What’s actually insecure is:
• Copy-pasted auth
• Hard-coded secrets
• Untested dependencies
• Security as an afterthought

Modern platforms are baking security into the workflow.

You don’t need to be a cybersecurity expert. I'm not!
You need guardrails that work while you build.

Shipping isn’t reckless anymore.
Shipping without visibility is.

Still building.
Still shipping.
Still letting the platform catch what I’d rather not.

𝗜 𝗱𝗼𝗻’𝘁 𝘂𝘀𝗲 𝗽𝗿𝗼𝗺𝗽𝘁 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀.
𝗔𝗻𝗱 𝗻𝗼𝘁𝗵𝗶𝗻𝗴 𝗯𝗿𝗲𝗮𝗸𝘀.

I keep seeing these graphics:

“8 Prompt Frameworks to Master AI”

 • R-T-F
 • S-O-L-V-E
 • R-A-C-E
 • D-R-E-A-M
 • Y-A-D-D-A-Y-A-D-D-A-Y-A-D-D-A
 • B-O-R-E-D 😁 

Every week I swear there’s new acronyms.

Apparently if you’re not using one, you’re doing AI wrong.

I don’t use them.

I just talk to AI like I talk to a colleague.

I explain what I’m trying to achieve.
We iterate.
We refine.
We adjust.

That’s it.

The biggest gains in AI output don’t come from acronyms.

They come from:

• Clear intent
• Context
• Constraints
• Iteration

You don’t need a secret formula.

You don’t need a laminated cheat sheet.

And you definitely shouldn’t feel anxious because you’re not using some viral template. 𝗬𝗲𝘀, 𝗜 𝗮𝗱𝗺𝗶𝘁 𝗜 𝗱𝗶𝗱 𝗳𝗲𝗲𝗹 𝗮𝗻𝘅𝗶𝗼𝘂𝘀. 

The whole promise of natural language models was this:

You don’t need special syntax anymore.

You don’t need 𝚙̲𝚛̲𝚘̲𝚖̲𝚙̲𝚝̲ ̲𝚎̲𝚗̲𝚐̲𝚒̲𝚗̲𝚎̲𝚎̲𝚛̲𝚒̲𝚗̲𝚐̲ ̲𝚒̲𝚗̲𝚌̲𝚊̲𝚗̲𝚝̲𝚊̲𝚝̲𝚒̲𝚘̲𝚗̲𝚜̲.̲

You just need to think clearly.

And with AI agents coming online?

You won’t even be writing prompts.

You’ll be defining goals.

So if you’re using frameworks, great.

If you’re not?

You’re not doing anything wrong.

You’re just having a conversation.

Quick build update on my multi-agent AI content system.


Added a few new pieces:
Strategy section
Content approval queue

improved 
Multi-brand dashboard
Content calendar
Queue health
Job monitor

It’s now running content for 4 different brands and showing:

• what’s planned
• what’s generating
• what’s ready
• what’s publishing
• what failed

Also built a small calendar widget so I can see exactly what’s going live tomorrow.

At this point the agents are doing most of the work.

I’m mostly just watching the dashboard and pressing the occasional button.

Official job title now:
Chief Button Supporter.

My mrs came home one day at 6pm to see me on the X-Box playing Fifa.

She laughed as normally i was at my desk till 8pm, 9pm...

the agents have definitely given me back some time!

𝗦𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝘀𝘁𝗿𝗮𝗻𝗴𝗲 𝗵𝗮𝗽𝗽𝗲𝗻𝗲𝗱 𝗿𝗲𝗰𝗲𝗻𝘁𝗹𝘆.
𝗜 𝘀𝘁𝗮𝗿𝘁𝗲𝗱 𝘂𝘀𝗶𝗻𝗴 𝘁𝗵𝗲 𝗯𝗿𝗼𝘄𝘀𝗲𝗿 𝗹𝗲𝘀𝘀.

And the terminal more.

This was a profound realisation.

I started my career in a terminal.
Decades later, I’m back in one.

Not out of nostalgia.
Out of behaviour.

Most of my work now starts in the terminal.

More and more with Claude Code. 
Experimenting with OpenClaw.
Sometimes inside Replit.

Intent. Command. Execution.

No dashboards, no onboarding tours, no feature archaeology.

That’s the pattern.

Every abstraction layer adds reach.
Until it starts stealing agency.

SaaS didn’t fail. It overshot.

For non-technical users, SaaS is perfect.
Opinionated. Safe. Finished.

For technical users, it’s starting to feel heavy.

Too many opinions.
Too many constraints.
Too many insane clicks to execute 1 feature. 
Too much distance between idea and outcome.

Coding AIs have changed the economics.

The setup tax collapsed.
The cost of bespoke 𝗱𝗿𝗼𝗽𝗽𝗲𝗱 𝘁𝗼 𝗻𝗲𝗮𝗿 𝘇𝗲𝗿𝗼.
Owning capability started beating renting tools.

𝗦𝗼 𝗵𝗲𝗿𝗲’𝘀 𝗮 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻.

SaaS will survive as a UI layer.
Not as the power layer.

Power users won’t churn loudly.
They’ll just quietly leave.

They’ll assemble their own stacks.
Loose scripts.
Agents.
Workflows.

Opinionated by the builder.
Not the vendor.

And the tell is this.

If people who lived through DOS, GUIs, browsers, and SaaS
are voluntarily going back to the terminal,

it’s not regression.

It’s leverage.

Three hundred rejection emails.

Not three. Not thirty. Three hundred.

That was me, early career, trying to get a foot in the door
at agencies that had already decided
what a marketer should look like.

Wrong background.
Wrong path.
Wrong box.

So I took a different route entirely.

I went deep into the systems side of things.
Computer science at Middlesex.
Industrial systems management at Westminster.
Then the deep end of enterprise , 
Nokia, Virgin, Accenture, QVC.

Not because it was glamorous.
Because it was where the complexity lived.

And complexity, it turns out,
is the best teacher you'll ever have.

While most marketers were debating creative briefs,
I was learning how global operations actually work.
How a GTM motion runs across 90+ countries.
How a global media budget behaves under pressure.
How brand decisions feel completely different
when the margin for error is zero.

It took years.
It wasn't clean.

But I noticed somewhere around 2017.

The agencies those people were working at
were starting to crack.

Not because of bad talent.
Because of bad architecture.

They were built for a world
where humans did everything manually.
Brief in. Creative out.
Report. Repeat.

That model doesn't survive
in a world where content demands are infinite,
channels multiply faster than headcount,
and data arrives faster than it can be interpreted.

So I stopped trying to join the old system.
And I started building a new one.

Growth Hakka became the lab.
Eight years of live experiments.
Real clients. Real budgets.
No safety net of a holding group behind me.

And then AI arrived, not as a novelty,
but as the infrastructure I'd always wanted.

Suddenly the thing I'd been trying to describe for years
had a working model.

Multi-agent systems that could research,
plan, create, and publish.
Not replacing strategy , 
executing it, at a scale no small team could.

JournoPitch came out of the same frustration.
PR shouldn't cost £5k a month in agency fees
just to get a press release written properly.
So I built a system that does it faster
and trains on how modern media actually works.

AICRM came from watching founders
drown in HubSpot complexity
when all they needed was a voice note
and a clear next action.

None of this was planned.
It was pattern recognition built from rejection.

The 300 agencies that said no
sent me somewhere far more useful , 
into the engine room.

Into the part of marketing most people
never bother to understand.

The systems beneath the surface.
The logic underneath the creativity.
The architecture that makes everything else repeatable.

What has this journey taught me,
boiled down to something useful?:

Marketing is an engineering problem
disguised as a creative one.

Most agencies were never built to solve it that way.

I had no choice but to be.

𝗠𝗼𝘀𝘁 𝗕𝟮𝗕 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀 𝗮𝗿𝗲 𝗸𝗶𝗹𝗹𝗶𝗻𝗴 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲 𝗯𝗲𝗳𝗼𝗿𝗲 𝗮 𝘀𝗶𝗻𝗴𝗹𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝘀𝘁𝗮𝗿𝘁𝘀.

Not because the content is bad.
Because the entire approach is built around the wrong objective.

Founders and marketing leads open LinkedIn,
post something polished and professional,
wait for leads to appear,
and wonder why the only people engaging
are other marketers doing the same thing.

Recent research shows that 77% of B2B buyers
complete more than half their research
before ever contacting a vendor.

Your buyer has already made a shortlist.
They've already formed an opinion.
And if you weren't visible, helpful, and credible
during that research phase , 
𝚢̲𝚘̲𝚞̲ ̲𝚠̲𝚎̲𝚛̲𝚎̲ ̲𝚗̲𝚎̲𝚟̲𝚎̲𝚛̲ ̲𝚎̲𝚟̲𝚎̲𝚗̲ ̲𝚌̲𝚘̲𝚗̲𝚜̲𝚒̲𝚍̲𝚎̲𝚛̲𝚎̲𝚍̲.̲

The average B2B LinkedIn strategy looks like this:

Post about company news.
Share a product update.
Publish a bland thought leadership piece
that could have been written by anyone.
Celebrate a new hire.
Repeat.

None of that builds pipeline.
It builds a digital brochure.
And nobody subscribes to brochures.

The deeper issue is a misunderstanding
of what LinkedIn actually is.

It's not a broadcasting platform.
It's not a press release channel.
It's not somewhere to shout about your features.

𝘐𝘵'𝘴 𝘢 𝘵𝘳𝘶𝘴𝘵 𝘢𝘤𝘤𝘶𝘮𝘶𝘭𝘢𝘵𝘪𝘰𝘯 𝘦𝘯𝘨𝘪𝘯𝘦.

And trust accumulates slowly,
through consistency, specificity, and genuine point of view , 
not through polished company updates
that read like they were approved by three committees.

I spent years inside enterprise organisations
watching enormous budgets go into brand-safe content
that generated precisely zero commercial tension.

Brilliant companies.
But the content? Institutional. Careful. Forgettable.

When I moved into B2B growth consulting,
I made the same mistake for longer than I'd like to admit.

I wrote what I thought a credible operator should write.
Professional. Measured. Inoffensive.

And it performed accordingly.

The pivot came when I stopped writing for credibility
and started writing from experience.

The messy bits.
The wrong turns.
The things I'd seen fail repeatedly across a hundred different clients.

That's when pipeline started moving.

Not because I got louder.
Because I got specific.

Ask three questions:

One, does the content address a real pain
your buyer is losing sleep over right now?
Not a pain you assume they have.
A real, specific, observable one.

Two, does it position you as someone
who has solved that pain before?
Not through credentials.
Through demonstrated thinking.

Three, does it give the right person a reason
to reach out, follow, or remember you?

𝗔𝘀 𝘀𝗼𝗺𝗲𝗼𝗻𝗲 𝘁𝗿𝘆𝗶𝗻𝗴 𝘁𝗼 𝗱𝗼 𝘁𝗵𝗲 𝗮𝗯𝗼𝘃𝗲 𝗶𝘁𝘀 𝗺𝘂𝗰𝗵 𝗵𝗮𝗿𝗱𝗲𝗿
𝗲𝘃𝗲𝗻 𝗮𝗳𝘁𝗲𝗿 𝗮𝗹𝗹 𝘁𝗵𝗶𝘀 𝘁𝗶𝗺𝗲.

If the answer to any of those is no , 
the strategy isn't building pipeline.
It's building the appearance of one.

This has been a tough realisation for me.

𝗠𝗼𝘀𝘁 𝗔𝗜 𝘁𝗼𝗼𝗹 𝗰𝗼𝗺𝗽𝗮𝗿𝗶𝘀𝗼𝗻𝘀 𝗮𝗿𝗲 𝘄𝗿𝗶𝘁𝘁𝗲𝗻 𝗯𝘆 𝗽𝗲𝗼𝗽𝗹𝗲 𝘄𝗵𝗼'𝘃𝗲 𝗻𝗲𝘃𝗲𝗿 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘀𝗵𝗶𝗽𝗽𝗲𝗱 𝗮𝗻𝘆𝘁𝗵𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗲𝗶𝘁𝗵𝗲𝗿 𝘁𝗼𝗼𝗹.

This one is different.

Last month, I built the same AI marketing agent twice.

Same brief.
Same inputs.
Same expected outputs.

Once with 𝙲̲𝚕̲𝚊̲𝚞̲𝚍̲𝚎̲ ̲𝙲̲𝚘̲𝚍̲𝚎̲.̲
Once with ̲𝙾̲𝚙̲𝚎̲𝚗̲𝙲̲𝚕̲𝚊̲𝚠̲.̲

I wasn't running a benchmark test for a blog post.
I was building a real component of my content automation pipeline , 
the multi-agent system I've been trying to acheive for months
that runs content across four brands and six channels daily.

So the stakes were real.
The time cost was real.
And the performance gap?

Genuinely surprised me.

I've been building in AI-native environments long enough
to have calibrated expectations.
25 years in marketing.
Systems thinking from a computer science background.
Fractional CMO work across SaaS, fintech, telecoms.

I thought I knew roughly what to expect.

I didn't.

One tool gave me clean, composable agent logic
that slotted into my existing stack
with minimal friction.

The other fought me at every architectural decision.
It wasn't broken.
It just didn't think the way I needed it to think.

That distinction matters more than most people realise.
Because when you're building agents , 
not just prompting them, actually building them , 
the quality of the tool's reasoning about its own behaviour
becomes the differentiator.

Not the feature list.
Not the pricing page.
Not the benchmark leaderboard.

How it handles ambiguity.
How it interprets your intent when the brief is incomplete.
How much it fights you versus how much it extends you.

There's also a compounding factor that nobody talks about:
cognitive load.

When a tool requires constant correction,
you stop building.
You start babysitting.

𝘈𝘯𝘥 𝘣𝘢𝘣𝘺𝘴𝘪𝘵𝘵𝘪𝘯𝘨 𝘢𝘯 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵
is one of the most expensive uses of a founder's time.

I tracked both builds properly.
Time to first working output.
Number of iterations to stable behaviour.
Integration friction with the broader pipeline.

The carousel below breaks down everything I found , 
including the specific moments where each tool either earned or lost my confidence.

If you're deciding which environment to build your next agent in,
swipe through before you commit.

taplio

The Battle

Fergus Dyer-Smith

Zohe Mustafa

Followers

Posts

Engagements

Likes

Performance ScoreZohe Mustafa's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Fergus Dyer-Smith

Zohe Mustafa

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Zohe Mustafa's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreZohe Mustafa's LinkedIn Performance