DEV Community: Vitalii Kiro

Vitalii Kiro: AI Customs Officer: The Only Honest Employee in Ukrainian Customs

Vitalii Kiro — Mon, 20 Jul 2026 18:43:57 +0000

Here is my new blog about AI agents for Ukrainian customs office. It was published in Ukrainian news agency site.

Imagine a customs officer who works without breaks or days off. Who can’t be called on the phone. Who can’t be slipped an envelope. Who doesn’t know your name, your connections, or who called yesterday from the "right" number. Who compares your declaration in seconds against thousands of similar shipments across the country, against world prices – and either clears it or raises a red flag.

This is what the AI future of Ukrainian customs should look like.

This isn’t science fiction. It’s something that needs to happen here and now. Not new open competitions, not public councils, not new fighters against old corruption.

It’s important not to turn this into another "digital hype" project from consultants in suits. Because real digitalization isn’t about convenience. It’s about changing the architecture of the state: when more and more decisions depend not on a person but on transparent rules – leaving less and less room for backroom deals.

That’s why this process needs to be led by AI engineers, not by former customs officials talking about new technology.

Three things AI guarantees that a human cannot

I’m not romanticizing the technology. AI has its limitations, and those need to be discussed too. But there are three areas where the algorithm’s advantage over a human is structural, not situational.

First: AI doesn’t take bribes. Not because it’s "virtuous" in some moral sense. It simply has no bank account, no needs, no fear. Calling it is pointless, sending it a "gift" is impossible. That’s not a virtue – it’s architecture.

Second: AI is consistent. The algorithm applies the same criteria to the first declaration and the millionth. It doesn’t know that this company is "ours" and that one is "theirs." It doesn’t know who called yesterday. Equal treatment isn’t a declared value here – it’s a technical impossibility of doing otherwise.

Third: AI leaves a trail. Every decision is logged. Every deviation from standard protocol is documented. If a human manually overrides the algorithm’s decision, it’s visible. That’s real accountability — not an oath on the Constitution, but a timestamped digital log of actions.

"But what about smuggling?" – where AI won’t replace a dog’s nose

Let’s be honest here. AI won’t replace a customs dog sniffing luggage. It won’t replace an officer who knows the local smuggling market from the inside. Physical smuggling – excisable goods, weapons, drugs – requires a physical response.

But it’s precisely where the state loses the most money that algorithms have the greatest advantage. Most budget losses come not from classic smuggling but from undervaluing customs value, manipulating commodity codes, and using shell intermediaries. This is exactly where an algorithm can automatically compare a declared price against thousands of comparable transactions and spot what the human eye often misses.

These are the billions that never reach the budget every year. And no investigation will recover them as effectively as a system that never lets them get lost in the first place.

What the system is afraid of – and why that matters

The biggest resistance to customs digitalization isn’t technical. It’s human and political.

"The flows" aren’t just corruption among rank-and-file inspectors. They’re vertically integrated schemes with interests at every level, top to bottom. Whoever "runs" a flow has no interest in a system where there’s no one left to call. So the fight for AI in customs isn’t just an IT project. It’s a fight against the specific beneficiaries of the current disorder.

That’s why, alongside the technology, three things are needed:

• an independent targeting center with direct access to data outside the departmental hierarchy;

• public analytics – regular dashboards of customs-value deviations by product category, a kind of "wall of shame";

• criminal liability for manually overriding algorithmic decisions after the fact.

Without these, AI will become an expensive toy that eventually gets "tuned" to serve the same people as before.

Orest Mandziy, whom the Cabinet of Ministers recently appointed head of the State Customs Service, came from NABU. That’s certainly better than yet another "insider" from the customs underground. But a detective, even the best one, is not a reform. If Mandziy genuinely wants to break the system rather than just replace its face, he doesn’t need to fight criminals by hand — he needs a new architecture in which crime becomes technically much harder, or impossible altogether.

That means one thing: digitalization and an AI-based risk-analysis system must become priority number one – not in a five-year strategy, but in the first months of his tenure. Every month without it is billions of dollars flowing where they shouldn’t. And no detective with a warrant will recover what a properly configured system could have simply never let out the door.

What AI can actually do at customs – and what it’s already doing

Let’s get specific. Because talk of "digitalization" often stops at pretty slides without getting to the substance.

Real-time risk analysis. AI simultaneously evaluates dozens of parameters for each declaration: weight, route, sender profile, atypical cargo characteristics, product photos, discrepancies between declared value and market prices. A human is physically incapable of analyzing that many parameters at once. An algorithm can, in milliseconds.

Anomalies invisible to the eye. Machine-learning systems can already flag inconsistencies in documents and detect patterns typical of undervaluation or product-code substitution schemes – even when each individual declaration looks flawless on its own.

Learning from its own mistakes. Machine-learning models can become increasingly accurate as data accumulates and algorithms improve. A scheme that worked last quarter ends up in the risk database by the next one. "Gray" operators can’t negotiate an "update" to the rules with an algorithm — unlike with the head of a regional customs office.

Who’s already on this road

This isn’t theory. While Ukraine hopes to eradicate corruption by appointing "one honest person," other countries are already building systems that no smuggler can slip a stack of dollars into — let alone a suitcase of euros.

China has rolled out Intelligent Customs Inspection, a system that uses AI to analyze scanned images of containers and luggage, in many cases substantially reducing the need for human intervention in cargo screening and selection for additional inspection. This is a country with one of the largest trade volumes in the world – and the system runs at industrial scale.

India is systematically deploying data analytics and machine learning in customs risk management. The technology tracks supply chains, profiles suppliers, and checks HS classification codes against actual goods – automatically, without a human intermediary at key stages.

The European Union is moving toward a "data first, form later" model: automated risk analysis is meant to replace the chaotic manual selection of cargo for inspection. AI solutions are already being used for tariff classification, fraud detection, and speeding up clearance of low-risk shipments.

The United States, through Customs and Border Protection, is actively developing AI support for supply-chain management, with an emphasis on detecting illegal goods and counterfeits through large-scale data analysis.

The World Customs Organization has set a strategic goal: moving toward SMART borders – Secure, Measurable, Automated, Risk-based, Technology-driven. Automated risk management is no longer optional – it’s the direction modern customs systems are heading.

Vitalii Kiro: AI hasn’t taken your job. But it’s already taking plenty of money

Vitalii Kiro — Tue, 02 Jun 2026 10:52:35 +0000

Sam Altman admitted he was wrong. The visionary who spent years predicting inevitable mass unemployment driven by artificial intelligence recently made a surprising statement at a Commonwealth Bank of Australia conference. He is "delighted to be wrong" about how quickly AI would eliminate jobs. Altman is now pushing back against the white-collar apocalypse that most Silicon Valley experts – including himself – once forecast.

It’s worth asking why the CEO of OpenAI suddenly turned so "pessimistic" about his own industry. The AI hype shows no signs of cooling, and plenty of office workers have already convinced themselves the technology is about to render them obsolete.

A quick terminology check

When media, investors, and managers talk about "artificial intelligence," they almost always mean Large Language Models, or LLMs. This is not intelligence in any meaningful sense – it’s a statistical engine that predicts the next word in a sequence based on trillions of examples scraped from the internet.

ChatGPT, Claude, Gemini – all LLMs. They don’t think, don’t understand, and certainly don’t feel. They generate plausible text. They do it well. And they do it at enormous cost.

True artificial intelligence – capable of learning, setting goals, and acting autonomously in an open-ended environment – remains either science fiction or a very distant prospect. The gap between "very smart autocomplete" and genuine reasoning is vast. AI startup founders tend to avoid this topic, since blurring the distinction makes it considerably easier to raise billion-dollar rounds.

Business is starting to sober up

Reality is beginning to bite. Chaac Pizza Northeast, one of Pizza Hut’s largest franchisees with over 110 locations on the US East Coast, is suing parent company Yum! Brands for $100 million. The franchisee was forced to adopt an AI dispatch system called Dragontail. Instead of speeding up deliveries, it slowed them down. On-time delivery rates dropped from 90% to less than half, and New York sales fell by roughly 10%. "Dragontail did the exact opposite of what it promised," the lawsuit states.

Starbucks, meanwhile, quietly pulled an AI inventory tool from all its North American stores just nine months after launch. The system couldn’t reliably count bottles of syrup. In the promotional video from the launch event, the scanner visibly missed a bottle of peppermint syrup. An omen, in hindsight.

Uber’s COO has publicly acknowledged that AI spending is increasingly hard to justify. "If you’re not actually able to draw a direct line between token usage and useful features shipped to users, that trade becomes harder to justify," he said — after the company burned through its entire annual AI budget in four months.

Microsoft is counting the cost too. In May 2026, the company began revoking Claude Code licenses for engineers across its Experiences & Devices division — the teams that build Windows, Teams, and Surface. The reason was straightforward: token-based billing had exhausted the annual budget well before the year was out.

A company that has invested roughly $13 billion in OpenAI and aggressively champions AI across its product lines couldn’t survive its own bill.

The scale of "optimism"

Research firm Gartner warned back in 2025 that more than 40% of agentic AI projects would be cancelled by the end of 2027, due to escalating costs, unclear business value, and weak returns on investment.

Despite this, Nvidia CEO Jensen Huang used GTC 2026 to propose giving engineers a token budget worth half their annual salary to drive more productive AI use. Under his framework, an engineer earning $500,000 a year should be spending another $250,000 on tokens. "Otherwise I will be deeply alarmed," Huang said.

The industry, meanwhile, keeps moving on momentum. According to Gartner’s latest forecast, global IT and AI spending in 2026 will exceed $6.3 trillion, with the AI sector alone accounting for more than $2.5 trillion. These astronomical figures keep climbing – with no guarantees of a return.

I don’t believe in the apocalypse. But I don’t believe in magic wands either.

AI is a powerful and expensive tool currently passing through what technologists call the "trough of disillusionment" – the phase that follows every peak of inflated expectations. We’ve seen this before: the dot-com crash of the early 2000s, the blockchain craze of 2018 that was going to "replace the banking system any day now."

Altman walking back his predictions, Pizza Hut filing lawsuits, Starbucks returning to hand-counting syrups – none of this is a collapse of the technology. It’s the normal process of a technology growing up.

Dismissing AI entirely would be foolish. LLMs are already delivering real, measurable value to doctors, lawyers, journalists, and engineers. That value is genuine, and the tools arriving in the next few years will be more capable still.

But panic and blind faith are equally unhelpful. While AI hasn’t taken your job, it is reliably taking enormous sums of money from everyone who bought the hype in bulk.

The difference between "artificial intelligence" and a "large language model" is the difference between a mind and a very clever parrot. Both have their uses. But nobody is handing the parrot the keys to the company just yet.

At least not right now.

Vitalii Kiro: The Drone War Is Over. The War of Algorithms Begins

Vitalii Kiro — Wed, 27 May 2026 11:27:30 +0000

When people say that Ukraine has become a “testing ground” for Western technologies, there is some truth to it. But that truth is incomplete — and even offensive. What is happening here has long outgrown the scale of someone else’s experiment. Ukraine is not merely testing borrowed technologies under combat conditions; it is independently creating technologies that will define the nature of future wars and the value of human life for decades to come.

More than 2 million hours — approximately 228 years — of combat footage have already been accumulated by one Ukrainian non-profit organization, which centralized video streams from over 15,000 frontline drone operators. Two million hours of real warfare — dust, smoke, camouflaged tanks, shattered roads, night attacks — are being transformed into one of the world’s largest and most unique combat datasets. And it is on this material that Ukraine is training its artificial intelligence.

Numbers That Change the Understanding of War

Let us begin with accuracy. According to estimates from the Center for Strategic and International Studies (CSIS) and Ukrainian developers, autonomous navigation and AI-assisted targeting can increase the effectiveness of FPV drones from 10–50% to 70–80%. The exact result depends on battlefield conditions and electronic warfare systems.
These are not theoretical assumptions but real frontline data confirmed in CSIS reports.

Yet dry numbers are only the tip of the iceberg. The real breakthroughs lie in the details.
The “Avengers” platform identifies enemy equipment in just 2.2 seconds. The algorithm was trained on a unique array of combat videos: tanks hidden in tree lines, armored personnel carriers stuck on muddy dirt roads, camouflaged artillery. No peaceful laboratory in the world could have assembled such a dataset even over decades. Ukraine obtained it in just three years of war.

The Griselda system operates on an even larger scale. It fully automates the interception and analysis of enemy communications: transcribing conversations, analyzing semantics, and building connection graphs between people, military units, and events. This reduces the need for manual analysis by 99%. The entire cycle — from signal interception to delivering ready intelligence into the Delta battlefield management system — takes around 28–30 seconds.

Half a minute from interception to decision-making. In modern warfare, this is enormous speed, although just a year ago the same process took hours.
Until recently, the concept of “cheap war” remained a theory discussed by military analysts. Ukraine has put it into practice. An FPV drone costing a few hundred dollars destroys equipment worth millions. Yet the key value now lies not in the drone itself but in its software. The advantage belongs to whoever adapts algorithms to new battlefield challenges faster. In this confrontation, software defeats hardware. That is why Ukrainian engineering teams updating systems directly in combat zones work far more effectively than traditional defense giants burdened by years of bureaucratic procedures.

Not a Testing Ground, but a Laboratory

There is a fundamental difference between a testing ground and a laboratory. On a testing ground, others’ hypotheses are tested. In a laboratory, your own are created.

Ukraine’s state defense-tech accelerator Brave1 has already registered more than 300 AI projects. More than 70 artificial intelligence and computer vision systems are already operating along the front line. After visiting Ukraine, former Google CEO Eric Schmidt invested more than $10 million into the military accelerator D3, calling the pace of local innovation “truly impressive.” Turkish company Baykar invested around $100 million into its own research and production center in Ukraine. This is the same Baykar whose Bayraktar drones became symbols of the first phase of the war in 2022.

People do not come here merely to exploit resources. They come here to learn.
The “Drone Line” project envisions the creation of a 15-kilometer unmanned kill zone along the front line, with plans to expand it to 40 kilometers. At the same time, the startup Swarmer is testing swarm technology that allows one operator to control multiple drones simultaneously.

The most important shift in this war is not the drones themselves, but the speed of the adaptation cycle. Classical defense industries implement new technologies over years. In Ukraine, this process has been reduced to weeks. Russians strengthen electronic warfare systems — Ukrainian teams rapidly change navigation algorithms. The enemy jams GPS signals — drones switch to computer vision for autonomous targeting. Enemy camouflage changes — developers instantly retrain target-recognition models. Ukrainian defense forces operate not like a traditional army, but like a dynamic software platform.

After Victory: What Will Remain

Once the war ends, the world will face enormous demand for two Ukrainian products. The first is obvious: combat experience unique in modern Europe. The second — far more valuable in the long term — is military artificial intelligence hardened in real combat.

Datasets, algorithms, architectural solutions, and engineering teams that developed systems under missile strikes will become a powerful strategic asset. Ukrainian defense tech has long reached the global stage. American company Palantir Technologies integrates analytics into Ukraine’s military ecosystem, German defense giant Rheinmetall is launching joint production projects, and European startup Helsing considers Ukraine a key platform for military AI development. American company Shield AI is studying Ukraine’s experience with autonomous systems and drone warfare. Brave1 has become the main gateway for cooperation between Ukrainian developers, NATO, and global defense corporations. The world sees Ukraine not as a theater of war, but as a laboratory for a new military-technological doctrine.

Ukraine is becoming the military equivalent of the legendary American DARPA (Defense Advanced Research Projects Agency). It was DARPA that created the internet, GPS, autonomous systems, and the first drones. Yet the United States spent decades funding universities and laboratories. Ukraine is building its own DARPA directly on the battlefield — under shelling, during air raids, and in constant confrontation with Russian electronic warfare systems. That is why Ukrainian innovations evolve faster: a technology either proves its effectiveness in combat or disappears within days.

But an important question already demands an answer today: who will continue developing these technologies in the future?

Education as a Strategic Shortage

Ukraine faces a paradox: we have achieved a unique technological advantage, yet we risk losing the people capable of capitalizing on it.
Students and school graduates usually choose traditional IT fields — software development, design, or marketing. Military AI, robotics, autonomous systems, and computer vision in defense remain niche areas. There is a lack of specialized programs, public demand, and understanding that this is precisely where national security intersects with successful careers.

Without at least a partial restructuring of the education system today, within 5–7 years Ukraine may inherit a unique technological legacy without the specialists capable of developing it further. In that case, the market will be filled by foreign talent, or Western corporations will simply acquire Ukrainian startups together with their teams and relocate them abroad.

This is not a theoretical threat but the natural logic of the global market. Ukraine is holding the technological front that will shape our future for the next twenty years. The only question is whether we will manage to educate a generation capable of inheriting this legacy.