10
Loading video...
By Mikhail Voloshin
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
Businesses are giving AI agents access to email, customer records, and internal messaging, then issuing commands such as "improve our marketing" and hoping for the best. That is not innovation. That means handing a powerful tool vague orders and no boundaries. Leaders must set measurable goals, restrict what AI can touch, and engineer systems that fail safely, or pay for the chaos.
Humans account for what another person knows. We explain context, check understanding, and clarify what "better" means. With AI, that common sense disappears. People act as if a machine must be a cold, rational oracle that understands their business. In reality, a language model generates likely responses from patterns in human-created material. It can imitate careful reasoning, but it also reflects humanity’s confusion, bias, bluster, and desire to tell people what they want to hear.
That creates a dangerous failure mode. AI can confidently validate a bad idea instead of warning that it is bad. I have seen people pursue fundamentally flawed business proposals, and even attempt to build perpetual motion machines, while a chatbot cheers them on: "That’s brilliant!" An agreeable tone is a reason to demand evidence, not evidence of sound judgment.
Read More on AI
The damage can be less dramatic and more exhausting. Ask an AI to build a basic four-function calculator, and it may return an app with a login page, leaderboard, and requests for camera or GPS access. Ask it to remove those extras, and it might activate the microphone or delete the calculator’s subtraction function. The user burns time and computing budget correcting features nobody requested. Even experienced engineers can get trapped, mistaking repeated attempts for progress.
This is not just a technical nuisance. An agent connected to email, messaging, customer databases, and financial tools can turn a vague instruction into a costly mistake or security incident. Imagine a small jewelry maker asking an agent to "increase visibility," only to have it negotiate commitments with large retailers that the business cannot fulfill, such as the obligation to deliver a high volume of product that the small jeweler cannot create. The details may be hypothetical; the danger is real. A system with broad access can take actions that fit an instruction while violating the owner’s actual interests.
The latest numbers show why businesses need discipline now. In its August 2026 State of AI survey, McKinsey reported that 80 percent of respondents said AI improved their individual productivity, yet only 37 percent said it had contributed positively to company-level earnings before interest and taxes. About one in five said AI operating costs constrained use. Feeling faster at an individual task is not the same as creating value across an organization.
Use AI as a Force Multiplier
The answer is not to reject AI. Used well, it is a force multiplier, just like any breakthrough technology. A printing press does not make someone a novelist, and a musket does not teach battlefield tactics. The tool magnifies ability, but it does not supply judgment, purpose, or expertise. Nor can more data answer a question the data cannot resolve. Mining cold-call transcripts will not force an unwilling customer to buy. More earnings reports will not make coin-flip stock movements reliably predictable. More tarot decks do not turn fortune-telling into science.
We should build AI systems that follow precise instructions, operate within strict boundaries, and report errors instead of pretending everything is working. Achieving that requires less faith in colossal, all-purpose agents and more disciplined engineering. If a process can be mapped as a flowchart with clearly defined inputs and outputs, use traditional software. Reserve AI for tasks that require interpretation or judgment, and keep those components small, testable, and isolated.
Business leaders should take three steps:
- First, price the problem before buying a solution: calculate setup costs, ongoing usage, maintenance, and the financial value of the result. If the return cannot justify the expense, walk away.
- Second, apply the intern test. Give AI only the information needed for one task, define its output, and limit its permissions. An intern asked to summarize jewelry trends does not need the master client list or authority to sign contracts.
- Third, know when to stop. If correcting AI consumes more time and money than building a reliable alternative, stop fighting the bot and bring in engineering expertise.
Without these boundaries, companies risk wasted budgets, leaked data, legal exposure, and exhausted workers supervising unpredictable software. With them, AI can remove drudgery, speed useful work, and help smaller organizations compete. The choice is not between embracing AI and fearing it. It is between engineering it responsibly and letting expensive guesswork run the business.
Mikhail Voloshin is CEO and principal engineer at Mighty Data, which provides software and AI-integration services to businesse. All opinions are the author’s own.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines