This whole series is about pushing AI further into your business, so let me balance the ledger, because the businesses that use AI best are also the clearest about where it stops. This is my working list of what stays human, not as a limitation of the technology, but as a design choice that pays.
Accountability
A decision needs an owner with skin in the game. AI can gather, summarise, draft, and recommend, and it should, but the signature is human. The hire, the price change, the "yes we can take this job", the tender submission. The moment "the AI decided" enters your vocabulary as an explanation, you have built an accountability laundering machine, and customers, courts, and staff can all smell it.
Design rule: AI drafts, humans decide, and the threshold where drafting ends and deciding begins is written down, not vibes.
The trust moments
Every business has moments where the relationship itself is the product: the complaint that could end the account, the bad news call, the negotiation, the client whose loyalty was earned over years. Automating these does not just risk the moment, it announces what the relationship is worth to you. Ironically, AI's best contribution here is upstream: clearing your calendar and briefing you thoroughly so that when the trust moment arrives, a well prepared, unhurried human shows up to it.
Design rule: the more a customer needs to feel heard, the faster the path to a person, and AI's job is making that person the best informed version of themselves.
Judgement under real ambiguity
AI judgement is genuinely good and improving, and I trust it with volumes of routine decisions daily. But when the situation is novel, the information is thin, and the downside is real, taking the dodgy job or not, trusting the new supplier, reading whether a key employee is about to leave, you want a human who owns consequences and can feel that something is off without being able to cite why. Pattern matching on the past is exactly the wrong tool for the situations that do not match the past.
Taste and standards
Someone has to define what good looks like: the bar for quality, the sound of your brand, the call on which work you are proud to ship. AI reproduces standards magnificently, mine now write themselves through every Project and system I run, but the standard itself is authored by humans and drifts the moment nobody is tending it. Keep a human editor in chief of your business, even when AI does most of the typing.
Care, where care is the product
If empathy is part of what people pay you for, health, advice in hard moments, anything where someone is frightened or grieving, then the empathy is not overhead to optimise away, it is the deliverable. Automate the paperwork around these moments ruthlessly, precisely so the humans have capacity for the moment itself.
The pattern
Read the list back: AI does the volume, humans do the moments. AI handles the hundred routine emails so a person can sit properly with the one hard one. That is not a compromise position, it is the actual productivity thesis of this entire series. The repetitive work was never the valuable part of anyone's job, it was just the part that filled the hours, and the businesses that win with AI are the ones that consciously reinvest the freed hours into the human column.
Write your own version of this list for your business, one page, and share it with the team. It answers the fear question from the previous article better than any reassurance, because it shows, in writing, what you believe people are for.
That closes the strategy block. Next, the fun part: concrete use cases you can steal, starting with the phone that never rings out.