01What to Learn Instead of Coding in 2027
If you are wondering what to learn instead of coding, Julian Goldie's answer is three AI skills: managing AI agents, mapping your workflows, and judging what comes back with a taste test. These are the skills people moved into once AI started writing most of the code.
The reason is simple. Google said AI now writes about 75% of its new code. So the work has shifted from typing every line to directing AI and checking its output.
In this episode, Julian shows each skill working on a real business job, the system he uses for each one, and a plan to start this week. You do not need to become a coder to build things. You need the skills the best engineers moved into.
- Agent manager: brief AI agents and correct their work.
- Workflow mapper: get your process out of your head and onto a page.
- Taste test: check every output before it reaches a customer.
02Why the Job Is Moving From Coding to Directing AI
The change happened fast. In October 2024, AI wrote about a quarter of Google's new code. By last fall, it was about half.
Now it is about three quarters.
The job changed with it. Engineers are now supervising AI agents. One big code migration got done six times faster than it could a year earlier.
The engineers did not vanish. Their job moved. There is less typing and more directing, checking and deciding.
Here is why that matters for a business owner. If the best engineers in the world are managing AI instead of writing every line, then you can skip the coding and go straight to the skills that matter now:
- Telling AI clearly what to do.
- Knowing the steps of the job before AI touches it.
- Deciding whether the result is good enough to send.
That is the core idea of this episode. The value has moved from writing the work to directing and judging the work.
03Skill One: The Agent Manager and the Four-Part Brief
Julian shows the agent manager handling a new lead from start to finish. He sends one request into Claude Code: handle this new lead. The lead is a bakery that opened a second shop, and nobody can find it on Google Maps.
First, the manager writes a brief for each agent:
- The researcher gets the lead's message and their website.
- The writer gets the tone, the services and the length.
- The checker gets the rules it has to check against.
Every brief has the same four parts:
- The goal.
- The context.
- What done looks like.
- What not to do.
Then the agents go to work side by side, and you can watch each card update as they finish. Next comes the part most people skip. The checker read the draft and flagged one line.
The writer had claimed this was a common problem for a new second shop, but nobody told it that, and it could not be traced to any source.
04How to Correct an AI Agent Like a Person
When the checker flagged that line, the manager did not write a whole new prompt. It sent a correction back to the writer with just the fix and the reason why.
The fix was to swap the made-up line for a plain one: yes, we can help with that. The rest of the draft stayed the same.
Julian points out that this is the same skill you would use to manage a person. You do not start their work over. You point to the problem, explain why it is a problem, and tell them what to change.
- Name the exact line that is wrong.
- Say why it is wrong, such as a claim that cannot be traced to a source.
- Give the replacement or the rule to follow.
This keeps your agents fast and keeps the good parts of their work. It also makes the agent manager skill easy to learn if you have ever managed staff. Clear briefs go out, work comes back, and short corrections fix what slipped through.
05Skill Two: Mapping Your Workflows Before You Automate
An AI agent can only follow a process that exists. In most businesses, that process lives in the owner's head. Before you hand work to AI, you have to get it out of your head and onto the page.
Julian gives the workflow mapper one job the business repeats every week: reply to a new lead. Then it interviews the owner with a short set of questions:
- What starts this job?
- What do you check first?
- What are the steps in order?
- What does a good result look like?
- Where do you decide, and what do you wait on?
The owner answers in plain words, and the tool draws the map. Every step becomes a card:
- The contact form email arrives.
- Read what they asked.
- Research their website and Google listing.
- Pick the one service that fits.
- Draft a short reply.
- Check it.
- Approve and send.
- Add the lead to the sheet.
Each card shows who does the step, an agent or a person, what it needs to start and what it hands to the next step.
06How the Workflow Map Finds Gaps in Your Process
On the map, the orange cards are the steps where a human still decides. That tells you which parts stay with you and which parts agents can take. The map then becomes the brief for your agents, so the agent manager knows exactly who does what.
The part Julian likes most is that the map finds the gaps. In the interview, the owner admitted that if a lead does not book a call, they usually forget to follow up. So the mapper flagged it and added a new step: follow up if no call is booked.
It found more gaps too, including:
- There was no clear status for when a lead is done.
- Nobody said where the leads sheet actually lives.
That is the real value. You do not just get a diagram. You find the holes in your own process before an agent ever touches it.
If you skip this step, an agent will follow the same broken process and repeat the same misses, only faster.
07Skill Three: The Taste Test for Every AI Output
Julian says this skill matters more than the other two. AI is fast, and fast means it can send something sloppy to a customer in seconds. Customers do not care that AI wrote it.
They care if it is wrong, sounds like a robot, or wastes their time.
So every output goes through the taste test before it reaches anyone. It asks three questions:
- Is it true? Every claim, name and number is checked against the source. If it cannot be traced, it gets flagged.
- Is it on brand? The draft is checked against your tone guide.
- Is it useful? It must answer what the customer asked, with a clear next step.
For the on-brand check to work, the tool needs to know how you sound. The first time you run it, it builds your tone guide. You paste in five things you have written that you are proud of.
The more you feed it, the sharper it gets.
This is also the skill no AI can do for you completely. You set the taste, and the system makes sure it is applied every single time.
08Results: The Bakery and Plumber Replies
Julian ran the taste test on the bakery reply from earlier. It passed the truth check, because the new shop's address and opening hours matched the bakery's website. But it was flagged on brand and came back with two fixes.
- He accepted both fixes and it ran again.
- It was still a little long, so there was one more fix.
- On the third run it was 47 words and passed all three checks.
Every fix he accepts gets saved into the tone guide as a lesson, so it does not make the same mistake twice.
Then all three skills ran together on a second lead: a plumber whose Google listing had six reviews and old photos. The agent manager opened the map for replying to a new lead and briefed the researcher and writer from its steps. The checker used the tone guide, with the bakery lessons, and caught two things straight away: too long, and the wrong wording for the next step.
One correction later, the reply was 45 words and passed all three checks.
The reply then waited for the owner to approve with one click. The agents never send it themselves.
09Mistakes to Avoid and a Five-Day Starting Plan
Julian lists the three mistakes he sees most:
- Automating a messy process. An agent just makes the same mess faster, so map it first.
- One-line prompts. Write my reply is not a brief. Use the goal, context, what done looks like and what not to do.
- Trusting the first draft. The check and the correction are where the quality comes from.
Here is how he would start this week from zero:
- Day one: pick one job you repeat every week.
- Day two: run the workflow mapper on it and answer the questions.
- Day three: mark the steps where you still decide and keep those.
- Day four: let the agent manager brief your agents from the map and run the job once while you watch.
- Day five: run the taste test on everything and fix your tone guide wherever it flags something.
By the end of the week, you have one job running without you, and then you pick the next one. Julian notes that the agent manager playbook, workflow mapper and taste test are inside the Agent OS in the AI Profit Boardroom.