I do not code in the traditional sense. At least, that is not how I think about what I do now.
What I do is brief agents.
I explain the business problem. I give the context. I describe the outcome I want. I set the constraints. I review the output. I correct the mistakes. I ask for another version. I test whether the result is useful in the real world. And increasingly, I see this as something a whole team can do, not one person experimenting with a chat window.
That may sound simple, but for a business operator it changes a lot.
In my last essay, Vibe Coding in a Suit, I wrote about the surprise of suddenly being able to build things that used to sit behind a technical wall. This is the next step in that thought, and the ground has already shifted under it. The industry itself is moving from vibe coding to agentic AI: from clever one-off prompts to agents that take a goal and carry out real work. This is my operator's view of that shift.
From Abu Dhabi, my interest is not AI in a demo. It is AI transformation inside real UAE businesses, where sales, finance, operations, dispatch, collections and reporting have to work every day.
The first feeling was surprise: I can build. The more useful realisation came next. The real skill is not using AI tools. It is learning to brief agents well enough that business intent becomes software, and then bringing those agents into the daily rhythm of the team.
What does it mean to brief an AI agent? To brief an agent is to give it a business goal, the context to understand it, the constraints it must respect and the review criteria for its work, so it can do useful work inside real company processes. It is different from asking a chatbot a question. You are not asking for an answer. You are asking it to do work.
Briefing is closer to managing than to coding
When I brief an agent, I am not asking for a quick reply. I may ask it to analyse a process, build a dashboard, review a sales pipeline, prepare a financial projection, track collections or monitor operational KPIs. It takes the goal and moves through steps: it reads the data, proposes a format, tests the logic, finds gaps and suggests improvements. The better the brief, the better the output.
And here is the part that surprised me. The skill of briefing an agent is very close to the skill of managing people. You need to be clear. You need to know what you want. You need to explain what matters and what does not. You need to be specific about the result, review the work, and correct it when it is wrong.
That is why I no longer think of prompt engineering or harness engineering as only technical skills. They are communication skills. If you can explain your idea clearly enough for an agent to build it, there is a good chance you can explain it better to your team too.
Why agentic AI feels different now
The new generation of agentic tools has become much better at understanding business context. Whether it is Claude Code, Codex, Cursor or the local agents arriving every month, the direction is the same: these systems increasingly grasp what the user is trying to achieve, not only what code needs to be written.
That matters for operators. Earlier, an idea had to pass through many layers before it became software: a business requirement became a technical document, then development, then testing, then changes, then more explanation. The original intent often got diluted somewhere in the middle. Now the distance is shorter. The idea can move from mind to screen much faster.
This is not a fringe experiment. Google Cloud frames 2026 as the shift from one-off prompts to agentic systems that run real workflows, and most large companies say they are already adopting agents. Microsoft's latest Work Trend Index describes the same move from the human side: as agents take on more execution, the operator's job moves toward setting intent, judgement and accountability. The direction of travel is clear, even if the road is bumpy.
It does not mean the operator becomes a software engineer overnight, or that IT teams become unnecessary, or that agents can be left alone with critical systems. It means the operator can now participate directly in the building, explore solutions earlier, test ideas and see rough working versions before anything becomes a large project.
The real value is AI agents inside business operations
When people hear AI coding, they picture software development. My experience has been broader. In a real-economy business, the value usually comes from improving the daily operating system of the company, not from launching a public app.
In practice, that looks like:
- Better dashboards and cleaner CRM views
- Sales pipeline visibility and stuck-deal tracking
- Financial projections and collection tracking
- Delivery and dispatch monitoring
- Internal KPI reports and operational follow-ups
- Exception tracking and clearer visibility for management
These are the tasks a smart analyst might do by hand. The difference is that an agent, working inside the right business context and with the right data protection, can support that work continuously. It does not get tired. It keeps checking performance and flagging where management attention is needed. That is not magic, and it is not perfect. But the best use of agents is not to replace thinking. It is to increase the number of useful things a business can monitor and improve.
Business-trained agents need a protected environment
There is one point I want to be very clear about. If agents are going to understand a business deeply, they need context, and that context can include sensitive data, private processes and strategy. So this cannot be casual.
The right approach is a closed, protected environment with proper data controls, internal governance and a strong IT team. We can use the best models available without treating business data carelessly.
A vibe coder in a suit is still not hands free. Implementation on the ground still needs people who understand systems, security, integrations, permissions and support. The operator defines the problem and drives the intent. The agents create, analyse and iterate. IT makes sure it is deployed safely. This is not humans versus agents. It is operators, IT teams and agents working together in a better loop.
Agents belong inside the team's workflow
The next step is not everyone opening a chat tool once in a while. Agents become far more useful when they sit inside the way the company already works, not outside it. That is where AI productivity turns into real business automation instead of experimentation.
The real shift is not from people to agents. It is from people using chat tools to teams working with agents inside the operating rhythm of the company.
The goal is enterprise productivity, not just individual task automation.
A sales team should not only ask an agent to write an email. It should work with an agent that understands the pipeline, the follow-ups, the stuck deals and the next action needed. A finance team should not only ask for a summary, but work with an agent that understands collections, projections and the reporting format management actually uses. An operations team should work with an agent that tracks dispatches, delays, exceptions and KPIs every day.
The agent stops being a separate assistant waiting for a question and becomes part of the operating rhythm. The team works with it, corrects it, improves it, and uses it to see the business more clearly. That does not make people less important. It gives them leverage.
The next productivity jump will come when every strong team member has agents working beside them, watching the process, preparing the next view and helping them act faster. A team with agents does not simply do the same work faster. It starts seeing more of the business, earlier and more continuously.
The Self-Improvement Protocol
Agents make mistakes. They misunderstand formats. They show raw data when I wanted a management view. They answer the question technically but miss how I actually want to see it.
So they need to be corrected in a structured way. This is the same idea I described in After the Prompt: The Operator's Harness as a Self-Improvement Protocol, or SIP: when an agent makes a mistake, the correction should not stay in that single moment. It should become part of how the agent works next time.
If a report comes back in the wrong format, we do not only fix today's report. We make the preferred format a standing rule. If it misses a KPI, that KPI becomes part of the standard reporting expectation. If it floods me with raw data, we teach it the difference between raw data and management insight.
The agent makes a mistake. We correct it. The correction becomes part of the operating method. The next output improves. It is not perfect, and some things need repeated correction, but the system gets better when the feedback is structured, in the same way a good employee learns from clear feedback.
Why operators have the advantage
A pure technical person may understand code better than a business operator. But the operator understands the messy reality of the business. The operator knows where the process breaks, which report is useful and which only looks nice, which exception causes pain, what the customer actually complains about, and which KPI is cosmetic versus which one changes a decision.
That knowledge is the scarce part. Agents are getting better at execution, but execution still needs direction. Define the wrong problem and the agent will solve the wrong problem very efficiently.
There is a quieter benefit too. Working with agents makes you a sharper manager. If your instruction is vague, the output shows it immediately. You quickly learn to state the outcome, give context, define what "done" means and separate the important from the decorative. That is exactly what teams need from managers. Not every operator needs to become a programmer. But every serious operator should learn how to brief agents.
We are still early
I do not want to make this sound more polished than it is. We are still early. Agents still make mistakes, still need review, still need security boundaries and a strong IT foundation, and still need humans who know when something does not look right. The work is not hands free.
The caution is real too. Gartner expects more than 40 percent of agentic AI projects to be scrapped by the end of 2027, often for unclear value, weak controls or runaway cost. That is not a reason to stay out. It is a reason to keep discipline. The faster the tools become, the more judgement matters, which is exactly why the humans who understand the business stay essential.
The old way was slow: idea, explanation, waiting, translation, development, review, correction, delay. The new way is more direct: idea, brief, agent, review, correction, improvement, implementation. That does not remove discipline. It raises the need for it.
From vibe coding to briefing agents
Vibe Coding in a Suit was my way of describing the first shock of this change: a business operator in the real economy suddenly able to turn ideas into working software. But the deeper idea was never the vibe. It was the briefing.
The suit matters because this is not only a developer story. It is a management story, an operations story, a business transformation story. I still do not think of myself as a traditional coder. But I also no longer think software starts only after I hand a request to someone else. The distance between an operational idea and a working tool has collapsed. My job now is to brief better, judge better, and stay responsible for what gets shipped.
That is why I say: I don't code. I brief agents.
Frequently asked questions
What does it mean to brief an AI agent?
Briefing an AI agent means giving it a clear business outcome, enough context to understand the problem, the boundaries it must respect and the standard by which its work will be judged. It is different from asking a chatbot a question: you are not asking for an answer, you are asking it to do work.
What is agentic AI in business?
Agentic AI in business is the use of AI agents that take a goal and move through steps to reach it, working inside real company processes such as sales pipelines, finance reporting, collections and operations, rather than only answering one-off questions.
Can non-coders build software with AI agents?
Yes, non-coders can build prototypes, dashboards and internal tools with AI agents. But production systems still need testing, security, governance and support from a strong IT team.
Why are AI agents useful for business operators?
AI agents help business operators convert workflow knowledge into practical tools. They can support dashboards, CRM analysis, financial projections, collection tracking, delivery monitoring and KPI reporting inside real business processes.
What is the role of IT if business users can brief agents?
IT remains critical. Business users can define the problem and guide the output, but IT teams are needed for secure implementation, integrations, access control, data protection and reliability.
What is a Self-Improvement Protocol?
A Self-Improvement Protocol, or SIP, is a structured feedback loop where an agent's mistake is corrected once and the correction becomes a reusable, standing rule, so future outputs improve rather than repeating the same error.
What skill should managers learn for the agentic AI era?
Managers should learn to communicate clear outcomes, provide context, define success, review AI output and build feedback loops. These skills improve both agent work and human team management.
How do teams work hand in hand with AI agents?
Teams work with AI agents by embedding them into repeatable workflows such as sales reviews, finance tracking, collections, delivery monitoring and KPI reporting. The agent supports the team continuously instead of acting like a separate chat tool.
Will AI agents replace team members?
The better framing is that AI agents give productive team members more leverage. People still provide judgement, context, correction and accountability, while agents help them analyse, monitor, summarise and improve faster.
Source notes
- Google Cloud, "AI agent trends 2026", for the shift from one-off prompts to agents running real workflows. URL: cloud.google.com/resources/content/ai-agent-trends-2026
- Microsoft, "2026 Work Trend Index: Agents, human agency, and the opportunity", for human agency moving toward intent-setting, judgement and accountability. URL: microsoft.com/worklab
- Gartner, "Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", 2025, for the discipline and risk point. URL: gartner.com
Keep exploring
Author note
Prateek Saxena writes about agentic AI, business operations and the practical use of AI agents in real-economy companies from Abu Dhabi. More on this topic on the AI agents for business operators hub. More in the media kit, on the Al Zaabi Group role page, or across the Journal.