Matt Watson

Serial Entrepreneur, Startup CTO, Podcaster, Blogger @ VisionaryCTO.com

I’ve spent over 20 years as a founder and CTO, starting in 2003 when I co-founded one of the earliest SaaS companies. We bootstrapped VinSolutions to $35M ARR and sold it for $150M in 2011.

Since then, I’ve built and scaled multiple companies. I’ve led engineering teams through every phase of growth. I’ve also learned firsthand what breaks when leadership, product, and engineering aren’t aligned.

In 2018, I started Full Scale. It began as a way to help myself scale faster with top-tier engineering talent in the Philippines. Today, Full Scale has over 300 developers supporting fast-growing tech companies that need to move efficiently and with clarity.

I recently launched Full Scale Ventures, a venture studio focused on solving real problems for startup founders and engineering leaders. Our first product is EngagePath, a tool designed to help thought leaders identify warm leaders from their followers.

Most recently, I wrote Product Driven, a book for engineering leaders who want to stop playing whack-a-mole with tickets and start building real product momentum. The book is about aligning culture, leadership, and customer focus. Product thinking isn’t just for product managers anymore.

I also host a podcast and write a weekly newsletter focused on the gap between business and engineering and how to close it.

Read the book and subscribe here: https://productdriven.com/connect

Learn more about Full Scale: https://fullscale.io

Founder & CEO

We started Full Scale to do software development the right way from the Philippines. Our 300+ employees are helping dozens of companies build their software quickly and affordably.

* Philippines based, English fluency is superb!
* Staff augmentation model, our team works directly for you!
* Focused on senior level developers
* No middle man project managers in the way

Startups don’t need a CTO. They need a CPTO. That’s a hill I’m willing to die on.

For 20+ years, I’ve had the CTO title. The only reason I’ve been successful is because I never acted like just the head of engineering. I acted like the Chief Product AND Technology Officer.

Building software isn’t about code. It’s about outcomes.

It’s knowing what matters to the customer.
It’s tying what you build to real value.
It’s iterating based on impact, not ego.

In the age of AI, this is even more obvious.

AI can write the code. That’s not the bottleneck anymore.

Product thinking is.

What should we build?
Why does it matter?
How does it change the user’s life?

The worst setup in a startup is a Head of Product and a Head of Engineering negotiating priorities while momentum dies.

Engineering and product can’t be separate empires.

They have to be fused.

If you think you need to hire a CTO in 2026, they better be a CPTO.

5x Founder & CTO | Author of Product Driven | Bootstrapped to 9-Figure SaaS Exit | CEO of Full Scale | Teaching Product Thinking to Engineers

Realizations you have after being a CTO for 20+ years:

1. Speed doesn't come from coding faster.

It comes from clarity, focus, and fewer surprise pivots.

2. Most engineers want to care about the product.

They just don’t get the context they need to do it well.

3. Vision isn’t an HR slogan or a slide deck.

It’s a tactical tool that guides real-time decisions across the team. 

4. The best engineers aren’t just builders.

They’re problem solvers. But only if they understand the problem.

5. Bad culture isn't loud.

It’s passive. Quiet meetings. No pushback. Lots of guessing.

6. “Done” isn’t when it ships.

It’s when it solves something for the user.

7. Teams don’t magically become product driven.

They become product driven when leadership acts like it matters.

8. Product debt slows you down more than tech debt.

No amount of clean code can fix the wrong thing built really well.

9. Focus is a leadership decision.

Not a prioritization framework.

10. If your engineers are stuck waiting for better requirements...

You don’t need a new process. You need to lead differently.

Learn dozens more lessons from my new book, Product Driven. Also now available on Audible!

Scaling Tech Teams through Product Thinking & High-Performing Offshore Talent | CEO @ Full Scale | Author Product Driven | Podcast Host

AI generated code is amazing until the shit hits the fan.
Production is down. Your best engineer is troubleshooting it.

There is one massive problem.

They have never debugged any of this code before. They didn't write the code, AI did.

They lack familiarity and muscle memory about how the code works.

It isn't much different than asking your engineer who started yesterday to go figure it out.

This is the ultimate problem with AI generated systems.

They work great until they don't. Then nobody knows how it actually works. There is no deep expertise.

How do we fix this?

Pray that AI can fix it when production is down?

undefined undefined

Author

Chief Executive Officer

Podcast Host

Founder, CEO, CTO

Here are the 4 golden rules of software development: 
 
#1 If it can be null, it will be null 
 
#2 The answer to every question is "It depends" 
 
#3 Getting users to actually use software is 10x harder than building it 
 
#4 Great requirements are more important than great developers

If you can build an entire application with a few AI prompts, so can your competitors.

You have nothing of unique value.

Real software isn't that simple.

Every new software project should start with the database, not the UI. 
 
The database is the foundation of most software applications. 
 
Bad database architecture can make it difficult to add new features later. 
 
The UI is simply the lipstick on the pig if the database is done incorrectly! 
 
Spending time understanding the UI and UX is really important. It helps shape the database architecture. 
 
But when it comes time to write code and build something, it always starts with the database.

Most “senior” engineers aren’t senior.
They’re just juniors who stayed at the same company for 5+ years.

They know every quirk of the legacy code.
They can tell you which bug comes back every quarter.
But ask them to design a brand new clean, simple service? Well...

Because memorizing a codebase isn’t the same as mastering fundamentals.
Surviving time-in-seat isn’t the same as developing judgment.

At Full Scale, we don’t leave seniority up to tenure.
We provide ongoing training so our engineers develop the skills that actually make them seniors:

⚡ Architecture thinking. How to design systems that scale and stay simple.

⚡ Problem-solving. Approaching new challenges instead of defaulting to old hacks.

⚡ Communication. Explaining tradeoffs clearly to teammates and stakeholders.

⚡ Product context. Understanding why the code matters, not just how it works.


Real senior engineers aren’t defined by years in the seat.
They’re defined by how they grow and how they make the team better.

With 300+ engineers, career growth isn’t optional.
It is critical to our success and to the success of the 85 clients we serve at Full Scale.

I need to hire offshore software engineers for my startup. 
 
Where should I hire them? 
 
How do I trust them? 
 
How do I make this work?

Our client just fired two local software engineers.

Not because they were bad engineers.

Because they refused to use AI.

This isn’t the first time a profession has faced a shift like this.

Travel agents who ignored online booking tools disappeared.
Stockbrokers who ignored electronic trading platforms disappeared.
Accountants who ignored spreadsheet software disappeared.
Journalists who ignored the shift to digital media disappeared.

Technology doesn’t wait for people to get comfortable.

AI won’t replace engineers.

But engineers who use AI will replace engineers who don’t.

Claude Code now commits 4% of all GitHub public commits. 

That is insane!

Every week I keep telling all of our team members and clients how important it is for them to learn to use these tools.

If you don't adapt to them, you are going to be left behind.

If you use it for a day and don't see the immense value in it, you are already left behind.

But is still like herding cats to get 300 people to use them and nearly 100 of our clients on board as well.

There will always be early adopters and late adopters.

The AI hype isn't hype for us. Claude Code and agentic engineering has permanently changed how we work.

4% of all commits is wild.

The real number of commits that used AI in some way is substantially higher.

Software engineering has a K-shaped skill gap right now.

The engineers embracing agentic AI are on the top curve. 

The ones who aren't are on the bottom, and the distance between them is growing fast.

We talk to leaders at tech companies every day. The things I'm hearing:
"I haven't written code since November."
"I'm getting 5x more work done."

Six months ago, a lot of us thought this was hype. 

Nobody thinks that anymore.

You're either figuring it out, or you're in denial. 

There's not much middle ground left.

This weekend we had "Wellness Day" for all of our employees.
A day dedicated to them and their families to relax.

Swimming, yoga, food, music, and lots of other activities for kids of all ages.

No laptops or Slack notifications. Just family fun!

At Full Scale, we believe that team work is more than sprints. We also need to slow down once in a while, and when we do, something special happens.

I learned a long time ago that without your team, you don't have a business. We have to make time to recharge, reconnect, and remember why we do this in the first place.

Thanks to all the Full Scale team members who made it out this year. Next up is the summer party in May!

Full Scale

The death to software companies is rarely the engineering team. It's the product team. 
 
Any engineering problem can be solved with money. At worst, it's a funding or management problem, not engineering problem. 
 
Want to build a nuclear bomb or put humans on the moon? With enough money... no problem. 
 
However, money can't solve a terrible product vision or business model. 
 
As they say... fortune favors the bold. In this case, a bold product vision. All too often, the problem is the lack of product vision. 
 
Many companies struggle with what to build, how to differentiate, and their target market. 
 
They meander around in circles with no strong direction. 
 
The engineering team is left rudderless and often lost in refactoring, maintenance, technical debt, and architecture rewrite oblivion. 
 
They produce no real product value to the end users. Instead of building cool products, they are left to focus on building cool code. 
 
Customers don't buy cool code. They buy cool products. 
 
Cool products require a bold product vision and focus. A startup can raise all the money in the world but can't raise a product vision. 
 
That usually comes from the founders and product team. 
 
If you want your startup to succeed, a strong product vision is the most important piece.

They were right about software engineering being dead.
Just not in the way people think.

Software isn’t code.

It’s architecture, tradeoffs, and understanding the customer well enough to know what not to build, and connecting technical decisions to real business outcomes.

That isn't dead and it isn't going anywhere.

That has always been the job. Most people just forgot.

If you think software engineering is just about code, your career is dying.

Don’t tell me you’re using AI for software engineering if all you’re doing is opening ChatGPT.

That was impressive in 2024.

It felt like a superpowered Google search. You could research faster. Draft cleaner code. Spin up a small function without digging through Stack Overflow.

But that’s the old baseline.

If you’re an engineer in 2026 and you say you’re “using AI,” the question isn’t whether you’re chatting with a model. It’s whether you’re running an agent with real context.

We’re not asking AI how to do something anymore.

We’re telling it what to build.

There’s a difference.

A quick prompt has no idea how your system is architected. It doesn’t know your naming conventions, your constraints, your edge cases, your tech debt, or the trade-offs you’ve already made. It’s guessing from the outside.

An agent operating inside your environment is different. It understands your repo. It sees your patterns. It knows the rules you’ve defined. Tools like Claude Code or agent-based Codex setups can take detailed instructions, apply them across a codebase, and execute within the context of your actual system.

That’s not a better Google search.

It’s a hundred times more powerful.

If your team says they’re “using AI” and all they’re doing is pasting snippets into ChatGPT, that’s like saying you’ve modernized because you use Stack Overflow.

Helpful? Sure.

Transformational? Not even close.

The shift is from Q&A to delegation.

From isolated prompts to persistent context.

From “how do I write this function?” to “here’s the objective, here’s the codebase, here are the constraints, go execute.”

That requires more from you, not less.

You have to define standards. Provide context. Be explicit about architecture and intent. AI is only as good as the environment you plug it into.

The teams that win over the next few years won’t be the ones casually dabbling with prompts.

They’ll be the ones who build systems around agents. Be that Claude Code, Cursor, ChatGPT Code, etc.

Are we sure this isn't a software engineer running Claude Code?

BBC news showed this 15 screen setup.

“Monitoring the situation”
“Matrix”
“Where is my mouse?”

No matter what your thoughts on this. I genuinely impressed and feel ashamed of my four monitor setup now.


I’m off to buy more monitors 🤣

Shouldn't a "public benefit corporation" be against mass domestic surveillance and autonomous weapons?

I'm proud of Dario Amodei for holding his ground.

I wouldn't want my technology used for those things either.

CEO and Co-Founder at Anthropic

NoSQL will destroy your startup.
It's actually the #1 reason all startups fail.

Because you’re missing the most important thing every business is built on.

The foundation.
The core.
The one thing you cannot fake, skip, or outsource.

You’re scaling infrastructure.
Refining your stack.
Debating databases.

But none of that matters if you ignore this one truth.

And by the time you realize what’s missing, it’s already eating away at your runway.

Here’s the twist:

I’m not talking about databases.

I’m talking about your funnel.

No SQL = No Sales Qualified Leads.

You don’t need a better backend.
You need people to sell to.

Because you can’t scale a business that’s never had real demand.

And no tech stack in the world will save you from no leads.

You don't need rockstar developers. You need rockstar requirements.

That was true in 2025. It's even more true now.

AI can write the code. It can't write the requirements.

Most of us are building really simple software. Display some data,
save it to a database, call an API. AI handles that now. A junior dev
with good prompts ships features in hours.

AI can't figure out what the customer actually meant.
It can't tell you which features matter and which are distractions.
It can't make the call when scope gets messy and someone has to say no.

The requirements, the clarity, the "why this and not that": that's
still a human job. Now it's the only job that matters.

AI made average developers rockstars.
It made bad requirements catastrophic.

"I used to think the code was the thing."

The most code-obsessed architect I know told me that yesterday.

Now he's watching four Claude Code agents churn out features while he figures out what to actually build next.

Code was never the thing.
What the code does is the thing.

Tech founders figured this out the hard way.
You can't sell code. You sell what the code makes possible.

Ready to harness the power of AI at your company, but not sure where to start?

I get it. There is a lot of information (and misinformation) out there about AI.

If you’re a CTO, CEO, or other company leader who is looking to leverage AI but needs some guidance, I've got you.

Our team at Integrity Inspired Solutions has been working with AI long before the arrival of ChatGPT. We’ve used machine learning for years to create solutions to help companies reduce employee burnout, scale their uniqueness, and massively outstrip their competitors.

Think of us as your trusted AI software partner.

We don’t just develop software for you, we work with you from day one to help you understand the AI landscape, identify problems, ideate solutions, and create a clear roadmap of next steps.

We’re not interested in applying AI for the sake of it or because it’s cool. We care about helping you use it where it can make an impact.

Our approach is human-centric (we apply AI in a way that helps your people do more of what they do best) and security-focused (no proprietary data leaks here). 

Ready to talk?  Click the "Book a Coffee Chat" link (in my featured section) to get started.

Philip Ledgerwood

The clearest indicator of AI-generated content isn't the em-dash or the rule of three or the "It's not X; it's Y" used every three seconds.

(Although that last one is actually kind of a good indicator, but that's not what I want to talk about.)

Last summer, I bought a book about AI agents.

If you're curious what the book is, DM me.  I'd say it here, but the author seems the litigious sort and already hit me up here when I left a review of the book on Amazon trying to get me to change it.

It was hyped up a lot on LinkedIn and mentioned case studies I was interested in.  Not many people were writing actual serious works about agents at the time, so I was pretty excited.

As I read the book, though, it was a slog.

Paragraphs and paragraphs of high level filler that meant nothing.

"When entering into the world of agents, strategy is crucial, and this is where security needs to come to the forefront.  The key is collaboration and a shared understanding of the vital spaces in your systems.  This is where real progress is made.  There are many layers to this.  Think of it like an onion--which also has layers."

That is not a direct quote from the book, but you get the idea.

Pages full of that stuff, and about every fifteenth page or so, there'd be an actual story or case study that was illuminating.

I ended up giving the book a three star review because there was some useful content.  Unfortunately that constituted about 20% of the text.

The rest was all about how "crucial" various abstract concepts were and analogies that explained nothing.

To me, my distaste for AI content has very little to do with all the em-dashes or overused rhetorical devices.

It's that it talks a lot and says very little.

Folks, I'm speaking from my heart here.

I would much MUCH rather hear your unpolished, unstructured, actual passions and thoughts and interests and perspectives than to hear your GPU tell me about how opportunity meeting alignment is where the real leadership happens.

I help companies use AI securely & purposefully | AI Consultant | Software Developer | Professional LinkedIn Gadfly

You know how people like to start LinkedIn posts with "Unpopular opinion:" and then follow that with an uncontroversial statement that most people agree with?

I decided I'm going to try that with the "It's not X--it's Y" pattern.

Except X can only be things that nobody would think it was in the first place.

Let's try it:

"OpenAI is poised to raise another $100B in funding.  It's not the Manchurian Dynasty handling the Taiping Rebellion--it's real investor confidence."

"Palantir is moving its headquarters to Florida.  It's not an octopus with inner ear problems--it's political strategy."

"The company behind TikTok is developing their own AI.  This wasn't just two hummingbirds getting to know each other--it was a clear market signal."

Does #AI actually steal source material?

I will answer your question with a question.

Have you ever heard of the KosherSwitch?  It's still a thing.

This is an invention that separates the act of flipping the switch from the actual turning on of your power.  Flipping the switch moves a bit of plastic, and then at a random interval, an internal sensor may or may not trigger the electrical charge.

Why would something like this exist, you ask?

The Torah forbids working on Saturday (yom shabbat), and in some of the more orthodox circles of Judaism, turning on power switches counts as work.

By disconnecting your action from the result, the switch technically allows someone to circumvent this being "work" because there's no connection between your action and the desired result.

Even within orthodox circles, this is a controversial topic, with some arguing that this is no different than having your power on an automatic timer and others arguing that this seems to be a very contrived attempt to skate under the letter of the law on a technicality.

When an #LLM is trained on source material - whether we're talking about art or books or whatever - it isn't actually storing that material anywhere.

Instead, it uses the material as a reference to build a map.

If it's text, it uses the material to map out what words tend to appear along with other words.

If it's an image, it uses the material to map out what pixels tend to appear along with other pixels.

In other words, it uses the source material to create a model of the patterns that exist in the source material.

But it doesn't actually store the source material or any references to it.  All that context is lost.  It has no cognizance that its model was informed by To Kill a Mockingbird or the Mona Lisa.

All that remains is the map.

"Oh ho!" you counter.  "But you can ask AI to generate an image in the style of Pablo Picasso or ask it for quotes from literature and it will produce them."

That's true, but the products are generated from a probabilistic consultation of its model.

In other words, there's no -direct- connection between the text of To Kill a Mockingbird and an LLM producing a quote from it.

It's -guessing- that the quote is in there.

Just like it's -guessing- that the image that it's generating is in the style of Studio Ghibli.

It just so happens that the model is so thorough that often the guesses are right, giving the illusion of it actually using the source material.

This thin disconnection between the actual source material and what an LLM generates is the legally questionable shield that protects these companies from a direct and actionable accusation of theft.

It remains to be seen how well this shield will hold up, but so far, it is.

taplio

The Battle

Matt Watson

Philip Ledgerwood

Followers

Posts

Engagements

Likes

Performance ScoreMatt Watson's LinkedIn Performance

Performance ScorePhilip Ledgerwood's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Matt Watson

Philip Ledgerwood

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);}Matt Watson's LinkedIn Performance

Performance ScorePhilip Ledgerwood's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreMatt Watson's LinkedIn Performance