13 October 2026
AI is replacing work. In some cases it is starting to replace workers. I do not think that is controversial any more.
What gets much less attention is that the same shift creates a new problem in the same breath. Every AI agent you put into your business creates something that has to be managed, and most organisations have not built the capability to do it. My thesis this week is that the scarce employee of the AI era might be the person who knows how to manage machines instead of people.
It is happening, but narrowly. I see it in specific roles and specific workflows, mostly at the junior and mid-level end, where there is a volume of effort that AI can genuinely take on. That is enough to make people rethink hiring, and they are.
What we do not have is wholesale occupation replacement. The reason is the one I keep returning to: judgment has not moved, and judgment tends to be crystallised in more senior people. So the impact lands at the junior levels first, which is exactly where you would expect it to.
If you want to think clearly about this, it helps to separate the job from the tasks inside it.
An occupation combines production and routine work, and that part AI handles well. It also carries relationships, exceptions, accountability and judgment. AI can take the predictable piece of work you do over and over. It is not capturing the rest of the human workflow around it, and I do not see that going away soon.
The obvious question is why an autonomous agent needs managing at all. The answer is judgment again, plus changing circumstances and constant tuning.
Look at what is already happening with agents doing go to market outreach. In organisations running that at scale, or that have really pivoted towards it, it is close to a universal experience that the agents keep needing tuning. The sequencing changes. The messaging changes. You do not set and forget an agent any more than you would leave your junior team members entirely on their own. They need intervention, and they need continuous thought and management.
Sidney put it to me that this sounds like people management 101, and it is close. You are still allocating work, measuring it and correcting it. The difference is that you also need to understand workflow design, data quality and governance, on top of bringing your own domain judgment to the problem.
Nobody has twenty years of agent management experience. If they claim it on their CV, you will spot it immediately.
There are two real reasons it is hard to find. The first is that the agentic systems are evolving so quickly that the skills have to shift with the platforms. Our own go to market tooling is fairly agentic, and we have changed it twice in six months.
The second is more encouraging. I think a genuine competence is starting to form around this. Our go to market engineering function was nascent twelve months ago, and now we can swap and change and adapt fairly quickly, because the capabilities we built in house have started to crystallise into consistent problem domains we know how to manage.
A lot like managing people. Agents need objectives, owners and boundaries.
Measure business outcomes rather than volume of activity. That is the same failure mode you see with people: plenty of businesses measure activity and mistake it for the outcome. And you still need to revert things back to a person. There has to be a human in the loop for it to work.
We live this at 1KE. The product generates solutions: it understands the customer's problem and builds a solution around technology. That is a complex problem domain, and even though 1KE produces the solution, we still expect the pre-sales architect to be accountable for it. It gives the architect scale. It does not replace the architect's judgment. The effort gets offloaded; the accountability stays.
At a workflow level, yes, and I think it is necessary. Agentic orchestration is already a thing, and without it you would have a million workflows all reporting back to one person.
What that removes is routine intervention. What it does not remove is the decision points where an organisation has to apply judgment, the same as we do in the analogue human world. Imagine an agentic workflow that does a lot of work and produces a quote. If the quote is low value and easy to quality check, maybe no human is needed. If it is an eighteen million dollar tender response, you probably want someone to look at it.
Well architected AI reduces effort. It takes some of the execution away. It does not take away the management or the judgment, and I think those become more important rather than less.
Alongside that, there is a new and emerging agentic management skill set that is genuinely rare right now. At an individual level that is an opportunity to build something valuable. At a business level it is a responsibility to foster it, and to think through what it means for how the organisation is going to run.
Watch the full conversation with Sidney on YouTube: https://youtu.be/LQAD7clWzC4
Subscribe to The AI Take for one AI lesson a week, and what it changes for your business.
1KE is the presales autopilot for technical sales teams. It turns discovery calls, emails, and technical documents into structured requirements and customer-ready solutions in minutes, so teams can respond while the deal is still warm.
Learn more → https://1ke.ai
Michael → https://linkedin.com/in/michaelchanter/ · https://x.com/1keceo
Sidney → https://linkedin.com/in/sidneyminassian/
Follow 1KE → https://linkedin.com/company/1ke-ai/
From the Blog
Ideas, frameworks, and updates for technical sales teams - delivered to your inbox.
