Author: Karthik Suresh

After three years I have started finally paying for AI. Here’s why.

I’m taking a new business to market with co-founders, focused on solving the data management problem in sustainability.

With our previous businesses we quickly reached a point where we needed more hands to do the work – there were more tasks to do than we had time for.

But AI can now handle these tasks more quickly and competently in an enterprise context.

But there are four principles for deployment that are worth remembering.

1. GIGO is still a thing

AI can do a lot, but it cannot unscramble your omlette. Give it good data, and it produces good output.

Give it garbage and it produces more work for you to fix.

1. Stay small to stay agile

It’s increasingly clear that you have to scope what you want carefully to get what you want. The bigger the scope, the less flexible it is.

Keep your projects small and contained. Then you’ll get the benefits of AI while staying agile.

1. Build what the customer needs

The time and costs of making stuff have gone down – unless you burn hours and tokens building things no one wants.

I’m spending 3-4 hours talking to customers for every hour of building – making sure I’m working on the right thing.

1. Build quality into the process

AI works so quickly that there’s no time to inspect everything.

Create outputs that are easy to validate. Build in checks. Make it easy to see what’s going on.

A good valdation process is the way to build trust.

As a new(ish) firm, we have the flexibility to start from scratch and build a consultancy that leverages AI.

But it’s not about cutting costs.

It’s about building a better way to do good work for clients.

It is a time of euphoria. Check if your pants are still on.

The Economist is puzzled this week.

• A jobs apocalypse may or may not arrive.
• Markets are correctly pricing or ignoring energy supply risk
• The S&P500 is on a rollercoaster, down 7% in April, up 8% year to date

How will we know if it’s real performance or not? Is this a bubble, or are we going to reach a top and then tip into recession?

We won’t know until it does. It’s very hard to predict the end of things.

Undertainty and optimism often coexist right before systems become fragile.

What matters is how prepared you are if the music stops.

As Warren Buffett wrote, it’s only when the tide goes out that you see who’s swimming naked.

Good risk management is never more important than when it seems everthing is going swimmingly well.

A company will not adopt an AI workflow just because it’s new and innovative.

We have to understand how people in the organisation make decisions.

In every conversation I have, we eventually have to address issues of power, politics and culture.

Who has leverage? Who wants what? How are things done around here?

You can’t get to commitment without first addressing these issues.

This is not a distraction.

This is the work.

The days of any one person being the expert in the room are over.

We all bring different types of expertise.

Consultants bring process expertise – this is how we’ve done this in other companies.

Clients bring situation-specific expertise – here’s what we’re struggling with right now.

A meeting may start with alignment – we’re here to talk about X.

But then it often diverges, as what you think we’re talking about and what I think we’re talking about drift apart.

The challenge is to bring us back together each time and realign the discussion.

This is actually the focus of my PhD research – where I’ve been developing a technique for managing and facilitating small group discussions.

We need tools that help us build shared mental models in real time.

Then we can agree on what to do next.

What you choose not to do may be more important than what you choose to do.

We’re often in situations where more information is needed before we can make a decision.

Should we apply for this grant? Should we build that capability? Is it worth buying this tool?

The real issue is the opportunity cost – are you making the best use of your limited time doing this kind of research?

But now, you can do a first pass with AI and then decide whether to invest more time.

Productivity is not about stacking hours of work.

It’s about making smarter choices about what to work on next.

Over the last ten years sustainability work has been heads down – collecting data, producing reports, and monitoring portfolios.

It’s time to go heads up – and show how climate intelligence can help manage risk.

A new generation of tools and APIs are making it easier to access high quality global geospatial climate data.

And spatial intelligence changes how decisions are made.

It lets us create customised risk models for large global portfolios.

Insights from those models inform how we manage risk.

For example, it’s one thing knowing that it gets hotter the further away you are from the equator.

It’s another knowing what percentage of your sites are exposed to extreme heat, and targeting them to create mitigation plans.

We don’t collect data for reporting alone.

We collect it because it’s essential to monitor, model and manage risk.

My job used to involve creating jobs.

We ran an innovation service, finding new opportunities, figuring out how to serve them, and then creating roles and teams to scale our service.

Now, I’m looking at each opportunity and asking myself whether it’s one for a person, or one for AI.

We should think of AI like we think of Excel – a general purpose tool that we reach for.

Old jobs were created when we needed to get something done, but didn’t have time to do it ourselves.

New jobs will emerge when we need someone to get something done better than we can.

That will undoubtedly include AI and automation.

Work is no longer about jobs and time on task.

It’s about delivering reliable, trusted outcomes.

In my first year of engineering school Rolls Royce came in to teach us how to build a jet engine.

A hundred students were divided into teams mimicking real factory departments.

I was in quality control. We told each team to creatively write the name of the part they were going to make on a piece of paper.

Then bring the part to quality control. We’d check it and pass it to the assembly team.

They put it on a board, and the job is done. We’ve built an engine.

Unsurprisingly, the first iteration was slow. Individual communication with teams took time. The artwork was great, but took time to create.

Then the task was repeated – do it again, but faster.

You can imagine how it went. Over a few iterations we improved communication and reduced the amount of work.

Less work. Fewer touchpoints. Quicker delivery.

23 years later, the Wall Street Journal has a piece on how Ford is going to compete with Chinese giants on car making.

They’re following this exact process.

Starting a skunk works to press steel in giant parts so that there are fewer components. Rebuilding the entire assembly system around limited touchpoints.

The takeway as a business/service/product builder trying to compete in 2026 against global competition?

Every extra handoff, approval and coordination step in your process costs you time and money.

Reduce the number of touchpoints – and keep it simple.

I once built a business that didn’t make me any money.

The problem was that I didn’t understand economics – in particular how prices are formed.

For example, it’s still not clear to me what will eventually set the price of the AI you use.

You can save time and money by having an AI build a financial modelling spreadsheet rather than having an analyst do it.

But is that what we’re really trying to do?

We don’t want to do the same crappy thing faster.

We want to stop having to do it at all.

As Ackoff put it – stop trying to solve problems. Dissolve them instead.

The real value from technology comes from redesigning systems so that unnecessary work doesn’t need doing at all in the first place.

Stop using AI to do the wrong thing faster.

Use it to build a better way instead.

Two things tell you all you need to know about a company’s strategy.

Its numbers, and its culture.

Greg Abel took over from Warren Buffett this year, and wrote this in his 2026 shareholder letter.

“… culture is a system for generating long term performance, not just a set of beliefs.”

Numbers are a starting point to analyze a business.

But numbers can be gamed and manipulated to present a rosy picture.

It’s much harder to hide the cracks in culture.

Numbers look back – telling you what a company did.

Culture looks forward – telling you what it’s willing to tolerate.