Author: Karthik Suresh

Organisations are socio-technical systems. Don’t forget the “socio” bit.

I’ve enjoyed a few days at the OR Society conference, learning from researchers pushing the edge of operational research.

AI hype is overwhelming and noisy, and we need to learn to ignore it.

Real organisations need much more than a network of AI agents.

Leaders have to set direction, work out which way the organisation is headed.

Managers and workers have to make sense of what is going on and what needs to be done.

And engineers and technicians need to build useful, practical, and sometimes novel solutions.

There is a prophesised world where AI will do all these things and people will be irrelevant.

We need to remember that right now, that world is imaginary, and is one of several possible futures.

Arguably, it’s not ever a terribly likely future.

But we have the world as it is now.

A complicated place, with complex problems.

And, as my kid’s rugby coach constanty reminds the team, let’s get on with the next job.

Which AI system is best for the majority of companies?

I’m tilting towards Microsoft and Copilot right now. Here’s why.

I see lots of people talking about Claude. It’s impressive. You can create complex and beautiful outputs in html with very little prompting.

It’s much quieter when it comes to OpenAI. It’s still very strong with text, but it seems to be stuck there – as a production tool for corporate speak.

I was searching for a metaphor to describe how I feel about these two systems.

Claude seems to me all hat and no cattle, as I believe the Texan saying goes.

It’s good for prototypes. But it’s a stretch to get from a demo to something that a company will use daily.

OpenAI is like a toupee.

It provides some cover, but its value is cosmetic.

Microsoft, on the other hand, is creating something really quite useful.

Copilot integrated in Excel, Word and Powerpoint is powerful and helpful. You can create complex models, turn transcripts into first drafts, and modify decks in unexpected ways.

Companies already have these tools, for the most part, so adoption friction is nonexistent, or minimal.

It’s a hard hat.

Something utilitarian and functional that does the job you need.

Yogi Berra once said, “It’s tough to make predictions, especially about the future.”

I’m optimistic about AI and its ability to make us more productive.

But there are issues that pop up from time to time.

We’ve already seen workflows break when a model changes.

I recently learned about the Lindy Effect – essentially, time is your friend.

If something has been around for some time, it will probably stay around for some time longer.

Corporations have used Microsoft for decades. Open Source is a foundational part of modern technology.

AI is still novel and changing rapidly.

AI lets you do things that weren’t practical before, but we shouldn’t confuse AI experiments with enterprise infrastructure.

That’s why we keep an eye on the latest technology but build our core services around Microsoft and Open Source.

Use new technology where it gives you leverage.

But build your systems so that they are understandable, maintainable and useful.

Why do tools that look good during demos fail to deliver in practice?

Donald Schön described this as the difference between the high ground and the swamp.

On the high ground people create methods and tools that are rigorous and research-based to solve manageable problems.

In the swamp are the messy and confusing problems humans have to deal with daily.

Tools that work for manageable problems struggle when facing messy ones.

SaaS demos look great with clean test data.

But they run into trouble when we try and use them to solve our particular problem in our particular context.

AI now makes it possible to create customised solutions that will do exactly what we need.

If they work, that shifts the complexity elsewhere.

Governance. Control. Reliability. Repeatabilty. Integration. Understanding.

The complexity doesn’t go away. We never get to drain the swamp.

Instead, we get to work on what needs to be done next.

Here’s what we really need from the systems we deploy.

They should make it easy to do the easy things.

And they should make it possible to do the hard things

The value of craft is in the struggle

Many years ago I went to a talk by a philosopher about the meaning of work.

After much research she had concluded, she said, that work was a task you wouldn’t do unless you were paid.

If you wanted to know what you want out of work, she argued, you first had to know what you wanted out of money.

I’m not sure that view translates to knowledge work.

Learning how to do something. Figuring it out. Finding a way. Struggling with the problem.

It’s closer to being a craft.

One that involves technology. We used to reach for Google to find information. Now we reach for an AI to give us the answer.

We’ve outsourced our memories. And now it’s very tempting to outsource our thinking.

But what we’re really doing is outsourcing the struggle.

The question isn’t whether AI can do the work.

It’s whether you want to.

It’s about the way in which you spend your days.

In 2026, if you want to know what you want out of work, you first need to know what you want out of time.

At the start of my career we prepped hard before every meeting.

We were ready. At the meeting we could talk for ages about what we did and the problems we solved.

We were the experts in the room.

But something was off. Prospects nodded in agreement and then went silent when chased.

It took me a while to learn that the problem was us – we were talking too much.

**The world isn’t full of clean, well defined problems waiting for your solution.**

Instead it’s full of messy situations that people find problematic.

Take something like working out a company’s carbon emissions. That should be straightforward, right?

Well, not quite.

… one company has a system but is struggling to use it.

… another is a large global enterprise, but is structured in a complicated way.

… another has changed methods three times and now needs to stick to one.

We can’t just rush into these situations and start offering solutions.

You have to first understand what’s going on.

So we slowed down. Listened more. Took notes. Asked questions. Drew what was going on. Checked that we understood what was in people’s minds.

What works? What doesn’t? What needs fixing?

Expertise helps us build solutions.

Understanding helps us build the right solutions.

The cardinal sin in operations is to do the wrong thing the right way

I’m reminded of this every time I work with a client to organise their sustainability data.

Real organisations have messy, complicated systems. They’re built over time. They’re acquired with new companies. They’re replaced, extended and modified over time.

All this is normal. It’s organic change, Small decisions made over time.

The problem happens when you try and take that mess and fit it into a preconceived system.

We know that any system has to match the complexity of the situation it’s trying to control.

Most don’t do that. Their world model just isn’t able to capture every permutation and combination of the real world.

So what works?

Building bottom up works.

Starting with the systems and data as they are, understanding what is where, and then pulling out chunks of data that can be organised.

That’s a flexible, extensible system that can adapt and grow as the organisation changes.

Because there’s one thing systems people have learned from bitter experience.

As soon as you get your system working, what it needs to do is going to change again.

In an AI-first firm safe roles either own context, or power, or both. All others are at risk.

We’re building a new consultancy from scratch, and my experience is very different from the businesses we’ve built before.

For example, you probably know which people in your firm do the actual work – the ones closest to the frontline.

In an information business, these people are the ones that own context – they know the client’s situation and what they need.

This used to be the preserve of less-experienced staff. That’s changing. Senior people are rolling their sleeves up and getting the work done because AI tools help them leverage their knowledge and dramatically reduce bureaucracy.

Leading from the back are people who are in charge of the business. They have power, either through ownership or because of the particular roles they hold.

They’re also relatively safe, because they make the decisions.

It’s the roles in the midde that will start to disapper, the ones that don’t own context and that don’t have power. The ones that can’t clearly articulate the strategic value they add.

A boutique firm hits the sweet spot. Small enough so the owners do the work. They own both power and context.

And the only thing that they now sell is value. Not time.

It’s time for prompt engineers to step aside. Context engineers are taking over.

Big consultancies have stopped talking about model options and now talk about the importance of judgment, defining workflows, and making activities predictable.

The technology at the centre will do what it is told.

But that’s not enough.

To do the right things it needs more information.

Is the solution just more contextual information – policies, guidance, guidelines and guardrails – or is there still a role for humans in an AI-driven workflow?

I think there is, because the real context is rarely explicitly written down

Instead it is constantly constructed, reconstructured and negotiated by the people who are involved in a situation.

If we don’t know what people want and need we won’t know what to ask the machines to make.

Humans are needed, first to understand the situation

Then to get machines to take the right actions.