AI didn’t cut the corners. Corporate culture did.
The week that was
In the space of a few days this month, Deloitte told staff it was cutting around 170 roles in Australia (some outlets put it at 200). In the same stretch, a significant cost benefit analysis Deloitte put together for merging Curtin, Murdoch and UWA for the WA Government was being called “AI slop” in WA Parliament.
Regarding the job cuts - Deloitte says the job cuts reflect a sustained shift in client demand. They also made comment on the less than favourable feedback on their analysis, saying that the errors in the report are minor and don’t change its findings. Officially, the two stories aren’t connected.
But this is the same firm that partially refunded a $440,000 federal report last year, after it was found to contain invented academic references and a fabricated quote from a court judgment. And it isn’t only Deloitte. This keeps happening, everywhere. The question I keep coming back to is why.
The easy answer
The easy answer is laziness. Someone pressed generate, skimmed it and hit send. Name them, tighten the policy, move on.
I’m not convinced that’s the whole story.
More than a few organisations I’ve worked with run on a powerful bias to action. Get it done, get it out, get it billed, get onto the next one. Speed is rewarded and throughput is measured.
In the same breath, we expect original thinking, deep engagement with the evidence and the kind of careful judgement that has always taken time. You can’t have both if nobody is given the hours.
The 30/40/30 shape
I think about good AI-assisted work as 30/40/30.
• The first 30 per cent is human. The question, the angle, the ideas worth chasing.
• The middle 40 per cent is where AI earns its keep. Gathering, structuring, drafting, packaging.
• The final 30 per cent is human again. Checking, questioning, cutting, refining. Deciding what’s true, what matters and what doesn’t belong.
That last 30 per cent isn’t proofreading. It needs creativity, intelligence and sophisticated reasoning, just as great work always has.
I’m not the only one thinking this way. Zambrero’s chief marketing officer, Bronwyn Heys, recently told B&T that marketers need to become a “human sandwich”: people set the direction, AI speeds up the middle, then people step back in to question and improve what comes out. It’s a great image, and I’ll admit it amused me a little, coming from a business built on convincing us to choose a burrito over a sandwich.
And that final 30 per cent, the one that demands creativity, intelligence and sophisticated reasoning, is the part that seems to be quietly disappearing.
Chipping away
There’s an idea usually credited to Michelangelo: sculpting is simply removing everything that isn’t the statue. Whether he ever said it is doubtful, but it’s too useful to let go.
That’s what the final 30 per cent feels like. When I review a draft over and over (and over) again, I’m not polishing. I’m chipping away everything that doesn’t support my point and make sure it lands with the audience I’m targeting. That’s where the all-important finishing touches happen, and where true quality comes from.
It’s also slow, unglamorous and almost invisible on a timesheet.
Close to home
I’ve seen this from people I’d stake my reputation on. Friends I regard as outstanding professionals, people I respect precisely for their attention to detail, have sent me emails with a tail of AI text still sitting at the bottom.
When I’ve asked, the answer is usually a sheepish “oh, that’s the AI bit I forgot to take out.”
These aren’t lazy people. They’re busy, capable people working in systems that have quietly stopped making room for the last 30 per cent. When quality stops being visibly valued, and the struggle for a great output stops being part of the job, people stop doing it. Not because they can’t, but simply because allowance hasn’t been made for the time.
So where does responsibility sit?
Individuals still own their work. “The AI did it” isn’t a defence, and nobody should be comfortable putting their name to something they don’t genuinely feel they authored, or that doesn’t properly make their point. “Will do” isn’t good enough.
But if organisations keep responding to AI blunders by punishing whoever got caught, we’ll keep getting the same result. The bigger lever, cultural change, sits with leaders and employers.
• Budget for the final 30 per cent. If review time isn’t in the scope, the fee or the deadline, it won’t happen.
• Make quality visible. Notice and reward the catch, not just the turnaround.
• Mind your leadership shadow. Leaders who send unchecked work give everyone else permission to do the same.
• Protect the struggle. Friction and dissent are often the birthplace of great thinking.
Thinking like a supervisor
It’s remarkable how fast this culture took hold. It offered more for less, and it spread like wildfire, fanned by the winds of profit and shareholder value.
But leaders who encourage it, chasing the dopamine hit of another closed loop and calling it productivity, aren’t thinking like leaders. They’re working to the time horizon of a supervisor, at best.
The organisational theorist Elliott Jaques spent decades studying this. He found that roles differ in how far into the future they carry responsibility. Frontline work runs to around three months, while the CEO of a major organisation might be working 15 to 20 years out.
The leaders worth following think in years, not weeks. The truly inspiring ones think in legacy. And what makes a business last sits a long way from closing loops: quality, innovation, trust, a reputation for work that holds up, and people who are proud of what they make.
Butter, not shortening
I want to share what’s happening with a couple of our clients.
After some digging inside their organisations, we realised their teams didn’t understand the true value AI could bring to the table. So we sat down with their leaders for a frank conversation about what AI is actually for. Not a faster route to the finish line, so people can move on to the next piece of research or the next serve of literary fairy floss. A way to lift the outcome itself.
You know the difference between a cake from a cheap bakery and one made properly. The cheap one looks the part, but it’s made with weird, oily shortening that leaves a film on the roof of your mouth. We’ve all grabbed one at a servo and regretted it by the second bite! The really special cake is made with beautiful French butter. It has substance, depth and a richness worth coming back for.
AI-assisted work can go either way. Rushed, it gives you the cheap cake: plenty of volume, and an empty feeling afterwards. Used prudently, with real thinking at both ends, it’s like the addition of golden French butter: the touch that makes the whole thing sing.
That’s the work we’ve been doing together: helping them see the market value, team engagement and stickiness that come from giving people the time to build something amazing. Done well, it becomes a sustainable competitive advantage. Done carelessly, it’s a race to the bottom, and right now it’s getting pretty crowded down there.
The 30 per cent that counts
AI is extraordinary at the middle 40 per cent: gathering, sorting, drafting and packaging at a speed none of us can match. But it’s the 30 per cent at either end that keeps the work original, considered and truly intelligent, rather than artificially so.
The organisations that prioritise that time won’t just avoid the next headline. They’ll be the ones clients pick up the phone to call, again and again.
Sources
• Deloitte to make 170 jobs redundant in outsourcing refocus, HRD Australia, 19 October 2026
• Tony Buti makes big splash with uni mega-merger, but now ripples are coming back, ABC News, 18 October 2026
• Western Australia university merger opposed by staff and students, Honi Soit, October 2026
• Deloitte to partially refund Australian government for report with apparent AI-generated errors, AP, October 2025
• Zambrero CMO wants marketers to become a ‘human sandwich’ or risk the scourge of ‘AI sameness’, B&T, 14 October 2026
• Time Span of Discretion, Paul Saffo on Elliott Jaques, Edge
• Elliott Jaques, Wikipedia