.when to use AI, and when to avoid it (and how I use it)

Today's newsletter strays from my usual psychological musings into something more practical - when to use AI, and when to steer clear.

I need to own the fact that I work with AI on a daily basis, although I try to steer away from being called an 'AI expert'. At Untrite, we build proprietary AI models that analyse real-time data streams to enhance decision-making capabilities - which sounds far posher than "we teach computers to read mountains of information at lightning speed while humans get to focus on actually thinking about them and taking action". Our platform integrates natural language processing and speech recognition to process diverse information sources - from police's emergency calls to internal documentation at corporations. By we I mean, our brilliant tech team does :) I wouldn't be able to tell you in detail about transistor architecture or break down the differences between BERT, GPT o1, and T5 models - that's what my business partner and co-founder Kuba is genius at (and he can easily translate tech and nonsense jargon into something that a non-techie decision maker can understand - a very rare skill!). My strength lies in understanding business problems and spotting opportunities - and that's why Kuba and I work in such great symphony. We marry technology with business sense to solve real-life problems. I know enough tech foundations to understand AI's potential and build upon it, while he masters the the deepest of deep tech.

I sometimes catch myself reaching for AI when facing topics that don't naturally excite me, but that I know are important to understand. That's why I often make myself write without it, usually on planes with no internet connection. Writing forces me to truly grasp the subject and form my own opinions. That mental wrestling match with ideas and verified, respected sources - that's where real learning happens (or so I hope ).

AI works brilliantly when you already know enough to catch its mistakes. But it won't make you an expert - that part's still on you. Take this post, for example. I wanted to write comprehensively about AI's usefulness, so I let AI handle the first draft to make sure I didn't miss anything obvious (to me). Btw, there's something deeply unsettling about people who publish AI-generated content without adding their own insight or effort. Are they even thinking about what they're saying? Do they actually understand the topics they claim expertise in? But enough of my light ranting. Let's get to the useful stuff.

The clever bits - what AI excels at

Remember when calculators were controversial in schools? People worried students wouldn't learn proper maths. Turns out, calculators are brilliant for complex calculations, but useless if you don't understand the underlying principles. AI's similar, it can be transformative when you know what you're doing, dangerous when you don't.

• Quantity over fatigue: Your brain gets tired after generating a few ideas. AI doesn't. And often, the best solutions come after exhausting the obvious ones or applying them into completely new fields. Want product ideas? Ask for 100, then have AI make them bolder and more unique. The key is specific prompting - vague requests get vague results.
• Expert's second opinion: GPT-4 solves PhD-level problems. Brilliant, until you realise you need expertise to know if those solutions are proper genius or properly mad. It's perfect when you're knowledgeable enough to catch its mistakes. Think of it as that wickedly clever colleague who's right 90% of the time but occasionally suggests solving inbox zero by deleting your entire email server.
• Translation champion: Got complex information that different teams need to understand? Since much of the current work on AI focuses around making machines understand human language, AI excels at translation between perspectives. It can make technical concepts accessible or complex ideas simpler, and translate you all that corporate jargon, technical specs and management speak.

How I use it

Remember when calculators were controversial in schools? People worried students wouldn't learn proper maths. Turns out, calculators are brilliant for complex calculations, but useless if you don't understand the underlying principles. AI's similar, it can be transformative when you know what you're doing, dangerous when you don't.

• Pulling essential info and next steps from business calls: I use AI to automate the soul-crushing bits of business that would otherwise demotivate me, like making notes during the calls - because I'd rather be properly present in meetings and focus on asking follow up questions to truly grasp the essence of a problem. With the number of meetings we hold, it's impossible to remember every client's particular flavour of chaos without decent notes.

• Breaking down those lengthy emails: some stakeholders love sending novels-worth of information in emails. AI helps me extract the actual action items and key points, making sure nothing important gets lost in these essays.

• Summarising documents: that's what we do for our clients, and naturally, we use it internally too. The fascinating (and often frustrating) thing about human language is its incredible flexibility - the same core information can be expressed in countless different ways. Traditional keyword-based systems would miss these connections entirely, failing to spot when two documents were essentially saying the same thing in different words. That's where AI's real power comes in - it actually 'understands' the meaning beyond specific words used. We've built Mira to do exactly this across broad customer service operations: helping teams in R&D, sales, and customer support cross-communicate effectively. It harmonises information across departments, extracts crucial insights from multiple formats, and creates a single source of truth from scattered documentation. AI can spot common threads that humans might take weeks to piece together. I use this daily, both for our internal operations and to demonstrate to clients how groundbreaking it can be when you properly unlock the value in your existing documentation (otherwise, you don't know what you don't know).

• Creating pros & cons tables when I'm too decision-fatigued to think.
• Converting general objectives into specific, measurable goals: you can't have vague, broad goals and hope you'll reach them without breaking them down into specific, measurable, (smart) mini tasks.
• Project proposal / RFP writing and refinement: these follow specific formats and requirements, so AI helps me and my team with the initial structure while I focus on customising the content to match exact client needs.
• Policy and procedure documentation creation and updates: keeping documentation current and consistent is crucial, especially when working with law enforcement clients. AI helps us streamline updates while ensuring nothing critical gets overlooked.
• Customer feedback analysis + automating responses: at both Untrite and my e-commerce venture Oishya, AI helps identify patterns in customer feedback and common issues that need addressing.
• Product description writing: particularly useful at Oishya for maintaining consistent, accurate descriptions while highlighting the unique features of each product. Also, I use AI for translations and syncs on the fly.

Creative:

• Making absurd AI-generated cover photos for this blog
• Background music generation for videos
• Voice synthesis for narration (using mainly ElevenLabs, Synthesia)
• Custom image and video creation for presentations and marketing (Midjourney, Pika, Runway)
• Video editing assistance (Capcut, Premiere Pro, Synthesia)

A serious note on security

I'm quite cautious about what business information goes anywhere near public AI tools. Client data, sensitive work details, anything confidential - that stays as far from AI bots. I'd rather not share the fame that Samsung got for using AI. For sensitive work, we either use our own fine-tuned models or keep it decidedly old school. Yes, it means occasionally doing things the slower way, but I'd rather spend extra time than explain to law enforcement clients why their confidential information might be floating around in a public AI's training data. That's the sort of career-ending move that makes accidental "reply all" emails look like a minor social faux pas.

When to stay away from AI

• The learning trap: Those AI summaries feel efficient but reading summaries isn't the same as wrestling with ideas until they click. Sure, I got through some literature classes that way, but did I really understand the books? No. Try using AI summaries for chemistry or maths and you'll quickly discover why that frustrating struggle with new material is actually essential. Your brain needs that resistance to build new neural pathways - it's not just about knowing the answer, it's about understanding how to get there.

• It sounds convincing: AI's errors are sneaky because they sound completely plausible. Studies show we get complacent, stop questioning because everything sounds reasonable. In situations where accuracy matters, trusting AI blindly is about as safe as using Wikipedia citations in your patent invention. You might get lucky, but when you don't, you may be in a big trouble. Just ask the lawyer who used ChatGPT to prepare court filings and ended up citing completely fictional cases.

• The failure patterns: AI doesn't fail like humans. Sometimes it becomes weirdly agreeable, confirming your wrong assumptions just because you sound confident. Other times it tries convincing you it's right with unshakeable certainty - even when it's completely wrong. The real danger isn't when AI is obviously wrong, it's when it confidently reinforces your existing biases. You need AI to challenge your thinking, not just nod along to every half-baked idea that crosses your mind.

• It keeps changing: there's no manual for AI's weird capability gaps. It might struggle with basic arithmetic but then write a perfect sonnet about quantum physics. These abilities keep shifting, requiring constant reassessment of where it helps or hurts. The solution? Test it thoroughly in low-stakes situations first. Never assume that because AI handled one task brilliantly, it can handle similar ones. Treat each new use case as a fresh experiment, and always have a human verification step for anything important.

The Real Value

The wisdom in using AI isn't just technical - it's paradoxical. It's most valuable where you're already expert enough to spot its flaws, least helpful in building that expertise. Think about those LinkedIn "thought leaders" publishing AI-generated content without understanding their topic. Are they really adding value? Or just contributing to the internet's largest collection of sophisticated-yet-astonishingly-similar sounding nonsense?

AI should amplify your expertise, not replace the work of building it. Everything worthy got to be hard to achieve. That's what differentiate the mediocre from the extraordinary. You need to put the effort. Use it to expand what you know, not skip the hard work of learning. Real expertise isn't just about having answers but about understanding why those answers make sense ...or not.