Jackbots of all trades: the problem with our pursuit of perfection

We’re at a point where lots of people are already unlocking the benefits of generative AI. Meanwhile, some leadership teams are still having conversations about things they could try, then getting stuck in a loop, debating potential benefits and barriers, endlessly iterating the perfect solution. This is often followed by a period of wondering why work doesn’t feel quite as transformed as they imagined, especially if they’ve blocked ChatGPT and bought a few Copilot licenses in a rush to demonstrate ‘grip’.

Why are we like this?

It’s a familiar scenario: people get excited about a simple solution and then — before putting it into practice and quickly unlocking the benefits they were excited about in the first place — they ask if it can also solve all their other challenges.

It’s because AI seems like magic, so we start to believe it can solve problems no product has ever solved before or create insights from data that doesn’t exist. Time passes while you become disappointed about things you can’t do (yet), and you miss out on the stuff you can be doing right now.

One of our innovation-savvy college customers made an observation that summed up the problem perfectly (possibly in response to a fleeting look of irritation during a meeting where our AI pro poker faces were not fully maintained). He said, “I think what we’re in danger of saying to you is: I really like this coffee machine, but I like to have toast with my coffee, so can you get it to make my toast too?”

Correct.

The other reason this happens is that we have an unconscious bias toward complexity. Follow me into this rabbit hole briefly. I promise it’s worth it.

When faced with problems, people tend to add elements rather than remove them, even when simplification offers a better solution with less effort. People aren’t doing it to be annoying: this is a hard-wired human cognitive bias.

It’s why I have to stop getting cross with the people with the forms and remember they’re just trying to add value — even if they’re taking things too literally and focusing on the ‘adding’ part rather than calculating the real ‘value’ they could create if they subtracted the work they made for themselves (and for me). This is how we are proven to function. Nature published a study.

That individual bias becomes endemic in workplaces. The concept of "peak complexity" suggests that societies and organisations tend to address challenges by adding layers of complexity, which can lead to diminishing returns and increased vulnerability. Over time, systems become overly intricate, making them less efficient and more prone to failure.

Understanding this bias toward complexity is essential — not just to help manage my red tape rage, but because it really matters when you’re trying to integrate AI assistants into workflows. Recognising our predisposition to add rather than subtract can help us design AI solutions that truly simplify activities and release efficiencies, adapting our work alongside.

Asking yourself what would happen if you didn’t add that new bit of bureaucracy and picking off problems and solving them one by one is the best way to avoid falling into this trap. Don’t try and fix everything, everywhere, all at once.

The myth of the mega-solution

Challenging unconscious bias and being prepared to park some unresolved issues that feel almost solvable with new tech is hard. That’s why we’re seeing a lot of organisations hold off on adopting quick, effective AI tools because they’re waiting for the solution. The one platform that integrates with everything, solves all their problems, and arrives fully compliant, user-friendly, and future-proof.

But the wait itself is costly. While you plan for the mega-solution, your teams are still stuck in manual processes, often overworked, and missing out on ‘quick win’ innovations. Worse, that “one platform to rule them all” may end up being a compromise: complex, expensive, and still struggling to connect with legacy systems — or locking you into a vendor relationship that’s more about control than capability.

The truth is that a relatively simple AI tool can do amazing things — especially when it’s trained or configured for a specific task.

We know that however we dress it up, a retrieval-augmented generation (RAG) system hooked up to your policy library does not sound glamorous, but it can radically improve decision-making, reduce risk, and free up your people to focus on more valuable work.

It won’t write your meeting minutes or build your dashboard. And that’s fine. You’re allowed more than one assistant. Teams Premium is great for meetings (and cheaper than going full Copilot then finding out it’s mainly just doing the minutes). You can get an HR assistant too, with its own resource library and training.

The toaster test

I love my new coffee machine: I got it in the Amazon Prime Day sale, and it’s changed my life for the better. I’m fond of my toaster too, and it does one thing, the thing it was designed and made to do, really well. They’re both simple, which means everyone else in the house can use them too, without my help or risk of injury.

I wouldn’t trade the quality of my morning routine for an all-in-one breakfast robot that will inevitably have multiple points of failure and does everything so-so but nothing well. My new coffee maker has not made, nor will it ever make, our 20-year-old Dualit toaster redundant. That thing will be toasting my hot cross buns for years to come.

The same goes for AI (and people, as it happens) in your organisation. Your data analyst doesn’t need to be your marketing copywriter. Your customer service assistant shouldn’t be managing policy compliance. You’re allowed a few best-in-class tools that each do one thing brilliantly — and often that means different picking suppliers for different things based on their specialisms. They tend not to mind if you’re shopping around. Horses for courses.

It’s also a sensible investment strategy: try different things alongside each other rather than betting the farm on One Big Thing.

What about agents?

AI agents will help — soon. They'll coordinate across tools, automate multi-step processes, and deliver a more joined-up experience. But they still work best when they’re orchestrating a well-chosen set of specialised tools. Even the smartest agent can’t make up for clunky systems, limited access, or a monolithic platform that resists modularity. Although you can start to see why your monolithic platform provider is incentivised to tell you that it can.

Microsoft Copilot and Google Gemini are evolving fast. They’re good at lots of things, and they’ll keep getting better. But they’re not the only answer — and they’re not the whole answer. They can be powerful parts of your AI toolkit, but only if you also make space for tools that solve your own, specific problems. Tools you can trust, tweak, and improve over time.

Start smaller. Get smarter.

AI doesn’t need to be big or expensive to make a difference. The most impactful tools are often the ones that are fast to deploy, easy to understand, and designed with a single purpose in mind. The real magic happens when those tools start working together.

They don’t need to solve everything to be worth your time. History tells us to make space for the innovations that are easy to overlook.

Did you know that Kodak invented the first digital camera in 1975? Their business model relied on selling film, so they didn’t commercialise it. Other companies got there soon after and didn’t wait. Digital cameras took over the market; Kodak filed for bankruptcy in 2012.

The lesson? Big organisations can seem unassailable in the marketplace, but it’s easy to reject game-changing innovations because they don’t fit the existing business or feel too “small” to matter.

Be more Shopify.

Shopify didn’t try to replace entire systems. Instead, they gave small businesses fast, focused tools to sell online. While big retailers invested in huge custom platforms, Shopify grew explosively by solving one thing well, and now powers major e-commerce players. Focused tools that solve a specific problem well can scale over time. You don’t need a mega-platform to create massive value.

‍