Every week, someone on LinkedIn explains to me why AI is not ready for production. They argue that it hallucinates. They point out that it makes confident mistakes. They insist you cannot trust it with real enterprise workflows.
They are not wrong about today. But they are catastrophically wrong about the trajectory.
The correct posture right now for every business workflow you own is simple. Add AI to the workflow. Keep a human as the approver. Be ready to remove the human when the next model makes it obvious you can. That is the entire strategy. The rest is just details.
Exponential Thinking Beats Gut-Feel Strategy
The nay-sayers are running a gut-feel model of AI progress. They tried a model in early 2024, watched it hallucinate, and concluded AI cannot do the job.
The reality is that AI capability operates on an exponential curve. What was impossible in early 2024 became routine by late 2024. What was unreliable in 2025 is production-grade today. What you cannot trust today will be trivial in eighteen months.
You do not plan a business for where the technology is this quarter. You plan for where it will be when your infrastructure choices finish compounding. If you put AI into your workflows now with a human checkpoint, you get two massive advantages:
- Immediate leverage: Even with a human review step, an AI-first workflow is faster than a human-only workflow. Every single time.
- Structural readiness: When the model gets good enough to remove the human entirely, you simply flip one flag. The cost structure of your business gets redrawn overnight. The competitor who waited to adopt is now twelve months behind you.
If you wait to adopt until AI is universally deemed "ready," you will find that the implementation ramp-up takes a year. By the time you finish ramping, the frontier has moved somewhere you can never catch.
The Mental Model: AI Types, Humans Judge
The cleanest way to handle the transition phase is to completely invert your old workflows.
The Old Workflow:
- A human enters the data.
- A manager approves it.
- The system processes it.
The New Workflow:
- The system enters the data via AI.
- A human reviews and approves it.
- The processing happens.
The human did not get removed. The human got promoted. They used to be the data entry clerk, and now they are the reviewer. You have the same number of humans, ten times the throughput, and dramatically less boring work. The humans are finally stationed exactly where they always should have been: at the judgment layer.
This is not a temporary productivity hack. It is a structural shift in what the human does all day. Furthermore, it is completely reversible. If the AI ships something dumb, the human catches it. The floor is never lowered.
Two Concrete Odoo Examples
Here is what this looks like in practice for workflows every Odoo consultant ships every week.
1. Sales Orders
- Old: A salesperson types the sales order into Odoo after a conversation with the customer. They make typos, skip fields, and forget to apply the correct pricelist. A manager double-checks at the end of the week.
- New: The customer conversation via email, voice note, or Teams transcript feeds directly into an AI. The AI drafts the sales order. The salesperson reviews it in one click, edits if necessary, and submits.
- The Result: The business output remains the same, but the "entering" step is gone. The salesperson's time collapses from twenty minutes per order to two minutes. The error rate drops because the AI reads the entire raw conversation rather than relying on a human trying to remember a phone call.
2. Requests for Quotation (RFQs)
- Old: A buyer manually fills out an RFQ in Odoo based on a consumption forecast, historical pricing, and multiple spreadsheets. They send it to vendors and wait.
- New: The system drafts the RFQ automatically from the forecast and historical data. The buyer opens a clean RFQ that just needs a quick glance and a confirmation click. They send it to vendors.
- The Result: The buyer's role has changed from data entry to a sanity check. The humans are happier, and the business moves faster. Nothing actually goes wrong because the human is still there when it matters.
The End State: Phasing Out the Human
Eventually, for specific and well-understood workflows, the human review will stop catching things. The AI will propose an RFQ, the human will click approve for the hundredth straight time, and somebody on the team will notice that you have not caught a real mistake in six months.
That is the signal. That is the exact moment you remove the human from the loop.
You execute this workflow by workflow, not all at once. Order entry might go fully autonomous before purchase approvals do. Payroll might take longer than inventory. Refunds might never go fully autonomous because the policy decision is the job itself.
This is a gradient, not a switch. But every single workflow on the path starts the exact same way. Add the AI, keep the human, and remove the human when the data says you can.
Your Action Plan for This Week
Pick one specific workflow in your business that involves a human typing the same kind of thing over and over. This could be sales orders, purchase orders, timesheets, expense reports, or customer support replies.
Put an AI at the front of it. Keep the human as the reviewer.
Measure these two exact metrics:
- The total time saved per transaction.
- The rate at which the human actually changes what the AI produced.
The first number is your immediate ROI. The second number is your countdown timer to full autonomy. Do not wait for AI to be perfect. Put it in the loop today so you are fully prepared when it is.