Direction · a layer of Growth Engineering
You automated the doing. Nobody is deciding.
Direction puts a person back on those calls: the machine handles its own routine mistakes and asks you only where judgement is genuinely required.
What the research shows
In a peer-reviewed study of 319 knowledge workers, researchers at Microsoft and Carnegie Mellon found that the more people trusted AI, the less critical thinking they reported doing. The effort moves from making the judgement to checking the output, and checking gets lighter over time.
The cost lands on someone else
Work that looks finished and is not gets passed along. Research from Harvard Business Review with Stanford and BetterUp found 40% of workers received this kind of work in a month, each instance costing about two hours to fix, and colleagues trusted the sender less afterwards.
The training gap behind it
The Marketing AI Institute has found lack of education and training to be marketers' top barrier for five years running, with 68% receiving no AI training at all. Teams are handed tools and no method, so the tool's defaults become the method.
Four levels, set by what is at stake
Routine work is done and logged. Anything that publishes, sends or spends tells you before it happens. A serious problem the machine cannot fix comes to you with the problem named first. Strategy, positioning and pricing are always yours.
A limit on how often you are asked
About three real questions a week, enforced by the system rather than by good intentions. Interrupt a person thirty times and they stop reading, which is how approval becomes a rubber stamp.
Your answers become rules
Say once that you will not pitch agencies under ten people, and it becomes a standing rule the system drafts under. We read it back to you before it counts, so nothing is invented on your behalf.
Both calls are scored
When you overrule the machine, your reasoning and how confident you are get written down next to its own. When the outcome lands, both are checked against it. Not as a scoreboard, but so the system learns where to keep asking and where to stop.
Everything we have written on this
Where this layer is put to work
- Demand creation: You choose the positions worth taking. The machine keeps to them and asks you only when a call is genuinely yours.
- Outbound pipeline: Nothing is sent without your yes.
Questions
How is this different from approving everything?
Approving everything is the failure we are avoiding. A decision that arrives with the flaw named, the options laid out and a limit on how often you are asked is a judgement. A queue of drafts to tick is not.
What if I am wrong?
Then it is written down, the same as when the machine is wrong. Nobody is scored against the other in public, and not stepping in is never held against you.
Will my team lose the skill?
That is the risk this layer exists to counter, and it is why the programme teaches the practice rather than the tool.
How much of my time does this take?
Minutes a day in Slack. The questions arrive in one place with what you need to answer them.
Does it ever act without me?
It never sends, posts or spends on its own.