Agentic AI is the revolution many business leaders hope for. But maybe it is something else- a replacement of a business.
In February 2026, Anthropic’s Claude was used to launch massive attacks on Iran’s soil. The results were devastating- its supreme leader was dead, and the war became the herald of the beginning of an AI-powered war.
This new and intelligent tech was on the frontier. A spy for the ages, one who could identify the patterns of information and suggest optimal paths for execution. And the business world used it as confirmation- the one with the smarter autonomous system wins.
Like all things computational, war is the grim foundation of proof. Proof that the technology works and must be applied to other areas, most importantly, business.
But AI agents are not prepared to run the economy, and let’s argue that they never will be ready. Any business leader who thinks the AI system is a replacement is either lying to themselves or others because they realize the value the perception is going to create.
Yes, AI systems will create value. It is a series of computations after all, and that is its greatest advantage and weakness. Like all things man-made, the AI systems are double-edged swords.
But, you will argue, humanity has always played with double-edged swords- what makes this one so particularly disruptive is the perception behind it: it will replace people.
tl:dr: it won’t. Even if the machines surpass people in terms of intelligence.
Let us show you why.
Agentic AI is not the replacement you’re searching for
Salesforce laid off almost half of its customer support roles and an additional 1000 in the february of 2026.
And the people laid off in the first round were intended to be replaced by AI. However, the program started failing- they had, in simple terms, overestimated what the tool was capable of. The leaders at Salesforce forgot that they were working at an enterprise, which is, by nature, human-run.
People are required to pass practical knowledge from one hand to another, make concessions, and anticipate outcomes based on their decisions- even while doing mundane tasks. This does not happen with agentic AI, not yet at least. Why? Because it does not know what the person knows- the stakes of it all.
It can assess your profit margin and assign an optimum path for its function’s execution, but once it starts to deviate, it deviates by a wide margin.
What seem like decisions and trade-offs are probability chains- ifs, if-else, buts, and other conditional operations, with a large margin for error, called hallucinations.
Of course, then come the AIs to monitor AIs or a self-recurring loop that improves on its errors: an AI to check the errors of another AI- an AI org chart! Which is already a reality.
It’s clear that while AI increases productivity and gains, it cannot decrease the rate of error or do tasks that require a kind of parallel thinking, because, by its nature, agentic AI, while thinking in branching possibilities, has linearity to it.
It cannot, like all machines, outthink itself. Living beings are gifted with metacognition, to think of 2nd, 3rd, 4th, 5th, and up to the nth order of effects. It comes to us as instinct or intuition. People pull from their experiences to solve problems and can anticipate the needs of the mission.
AI systems, on the other hand, are trained on very specific training data, which one should always assume has a lot of errors and gaps in it. What happens when these errors are reinforced or when outliers are deemed the norm?
These AI systems will face a massive and exponential failure rate- one that a sophisticated hacker or malicious actor can exploit.
The Human-AI-Organization Dependence Continuum
But AI and autonomy are inevitable- no organization is going to pass up this golden opportunity. However, the human, especially juniors, must always be in the loop.
Why?
Because continuous learning, by definition, is ongoing. Why does an organization, in a capitalistic economy, wish to expand? Stagnation means less innovation, fewer opportunities for growth, and low-value creation for shareholders.
Organizations are dependent on people for growth, and people are dependent on AI for automation, and AI is dependent on people to learn new avenues of problem-solving.
A positive loop! But bloated organizations are inefficient- it is not just a business problem but a social one. AI cannot replace organizational inefficiencies, which are plaguing the market, nor can it escape the grasp of competing self-interests.
It would be a folly not to take these vectors into account while making decisions for your organization.
Who takes the responsibility for agentic AI?
Think of this: you have sensitive client data, and because of a malicious prompt injection, API pull, or any other conceivable pathway, your AI system is compromised, leaking the data.
Who takes the fall but you? To prevent this scenario, a human must be in the loop at all times to act as a safeguard and to catch this from happening. Accountability is a vital component of any organization. And when there is no accountability, the default falls on the largest shareholder.
It opens up a singular entity to failure and to reputational disruptions. In the case of Salesforce, it has created a nosedive in its shares.
Agentic AI and SaaS: Is this assured destruction?
We can chalk Salesforce’s drop in shares to the SaaS market dip- but the investors are right here. AI is going to disrupt solutions that are knowledge consolidators with a nicely packaged UI/UX.
Querying an AI to consolidate knowledge will be easier, especially if it can create files- Claude’s file creator is an excellent tool.
But, you ask, what does that have to do with you not replacing your teams with AI agents?
What this is proving is that AI is also just a software that consolidates knowledge and executes a command. Yes, it might replace SaaS- that is its real threat- because at the end of the day, AI, even a super intelligent one, is a tool that will perform a function.
Whether that is business handling or attacking another nation, it will move in the direction of the user, a human with pre-determined goals. Or, Terminator is going to be very real very soon.
While the example may seem like an overkill (it is.), it clearly illustrates the spectrum these thinking machines are on- and on every spectrum sits a human being deciding where this intelligence needs to move- of course, unless AI achieves complete autonomy. Then it’s bye-bye.
Possible end of SaaS
But agentic AI is coming for a particular type of job- the SaaS industry- SaaS has been comfortably dominating since the 90s, or rather, software to be exact. Everyone wants a start-up, to be a marketer at a software company, and to see their company go public- well, that was before AI came and started eating away.
A lot of jobs were created because software development required it, but what happens when the fundamental basis of software changes or its requirements? That is the question that is left unanswered.
What happens when the tools you have built become obsolete overnight? Scary, isn’t it- that’s the same type of fear many people deal with when leaders say that their roles might become redundant.
So what does come next? Thinking of what we can do with saved time- either for progress or to fill the coffers of an already wealthy shareholder.
