Machines can say anything – but only humans can say something.
In a previous piece, I argued that AI is now excellent at generating “dots” – fragments, drafts and patterns – while humans bring chaotic energy, judgement and consequence. We carry the messy context. We live with being wrong.
So if machines can generate almost infinite dots, what does that mean for how we actually structure work, teams and careers?
Rethinking delegation: what machines should do vs what humans must do
If we pretend nothing has changed, we end up with a brittle structure: a handful of senior people doing all the real connecting, with everyone else feeding prompts into tools and polishing outputs.
A more honest response is to redesign the division of labour:
- Machines get the repetitive pattern work.
- Wherever the task is high volume, rule-bound, format-driven, the model should be your first instinct.
- Humans own the ambiguous decisions.
- Anywhere the task touches ethics, politics, long-term positioning, or real-world fallout, humans stay in charge – and AI becomes a sparring partner, not the decision-maker.
Everyone moves into “connecting work” earlier.
Juniors shouldn’t spend three years doing tasks a model can now do in three minutes. They should be learning to ask better questions, build sharper narratives and argue for/defend their choices.
The differentiator then becomes how a person frames the problem and what connections they are willing or able to make that others can’t.
Turning chaos into a method: The Creative Codex
A few years ago, we started formalising how we do that at Hoffman. The result became a methodology I now call the Creative Codex – our internal logic engine for campaigns and platforms.
The Codex does two big things:
- It forces every creative idea to pass a logic check: does the narrative actually follow from the problem, or are we just jumping to a cool activation or slogan?
- It self-checks for contradictions: if the tension says one thing and the “big idea” quietly assumes another, the Codex will highlight that gap.
The way I think about it: the Codex treats creative development a bit like writing code. If the logic doesn’t compile, it doesn’t run – no matter how good the headline sounds.
More recently, we turned that methodology into a set of custom GPTs that sit inside our workflow. They don’t spit out random ideas; they behave more like a logic harness for creativity by:
- Running through a far more complete set of troubleshooting methods than a human usually has time for when an idea isn’t sharp or a creative route isn’t flowing
- Pushing us to tighten the logic before we start decorating it with headlines, visuals or taglines
- Flagging fuzzy leaps in the story so we can fix them while it’s still cheap to change
- Taking the strongest parts of a “broken” idea and walking the logic both forwards and backwards, then recompiling it – often revealing alternate ways to solve the brief that we wouldn’t have found on our own
In practice, that means we’ve learned to use AI to hone chaotic energy, not erase it. Humans bring the leaps of faith – the random connections that feel like they might belong together, the jump from one piece of logic to another without a safety net. Then we let the Codex and our custom GPTs do what they’re better at than any of us: run the logic. They try endless ways to rearrange the dots, adjust the order, and smooth the joins until the story actually flows.
But there’s a catch. When AI is done flowing the logic, it usually sands off the edges that made the idea interesting in the first place. That’s where human chaos kicks back in. A seasoned powerpointer creative looks at the neatly formulated answer and starts breaking it on purpose – eliminating the safe bits, adding back tension, reconnecting only the dots that make the words sing like poetry.
That is chaotic energy in session. That is what it means to design for productive chaos. It’s accepting that chaotic energy is a sacred asset in the work, and treating it with intention instead of leaving it to chance.
A few practical moves
- Plan for weirdness – and leave room for wandering.
- Factor in unconventional thinking time where the goal is to think differently, not just think more – in unusual environments like walks, cafes, train rides or “slow meetings”. At the same time, protect pockets of the day that are deliberately unplanned. Creativity is rarely conventional. Your methods shouldn’t be either.
- Encourage cross-contamination.
- Mix people from different markets, disciplines and seniority levels on the same brief. Chaos thrives on unfamiliar combinations.
- Use AI as a provocation engine, not a vending machine.
- Instead of asking for “the answer”, use it to generate alternative framings, analogies, objections. Then switch it off and let humans fight it out.
- Make your frameworks visible.
- Codify how your best people think – into playbooks, canvases, or AI assistants. The goal isn’t to make everyone identical; it’s to give more people access to your sharpest questions.
- Reward the connection, not just the output.
- In reviews, spend time unpacking how the team arrived at an idea – which dots they chose to connect, and what chaos they embraced or ignored. The more you understand how the sausage was made, the easier it is to remember the ingredients and recipe next time.
Roads? Where we’re going, we don’t need roads.
AI will get better at connecting the dots. More compute, better models and richer feedback loops will keep pushing the frontier. Some of what feels distinctly “human” today will quietly become table stakes.
Right now though, in the actual budgets and tools that show up in comms and marketing teams, we are still in a world where:
- Machines are extraordinary at generating patterns and options
- Humans are still on the hook for framing, judgement and consequence
That in-between phase is where the risk and opportunity sit.
Leaders who treat AI purely as an efficiency play risk sanding off the very chaos that makes their stories, strategies and cultures resilient. Leaders who romanticise chaos and ignore the machines risk burning out their people doing work a model could handle in seconds.
The interesting path is in between: codify your best thinking into systems and tools; let AI handle the heavy lifting inside those guardrails; protect and train the human chaotic energy that decides which dots count.
At Hoffman, that’s the philosophy behind the Creative Codex and the custom GPTs we’ve built around it. We’re not trying to tame chaos out of existence. We’re trying to give it enough structure to do its best work.
If you want to explore how this approach might shape your next campaign, market entry or visibility push, get in touch with your local, very human, Hoffman team.
Machines can say anything – but only humans can say something.
