gshc2020.com

Organizations are racing to deploy AI agents across every function they can find. The appeal is easy to understand—instant scale, tireless execution, a workforce that never sleeps—but too few are stopping to ask how much debt they are creating in doing so. ​

This is not technical debt in the old sense, though the two are close cousins. Agent debt is the accumulated cost and complexity created when AI agents are deployed without a shared architecture for communication, context, identity, delegation and governance. Every agent added without that architecture becomes a disconnected piece nobody is watching. Like any debt, it grows until it comes due.

I have spent more than 30 years in technology across consulting, sales and product development. As founder and CEO of Saguna Consulting, I work with enterprises deploying AI systems through this exact shift. That is why this problem sits close to home, and how I know that the hardest part is understanding what it actually costs an organization to leave each of those five gaps open.

Debt #1: Communication Is Where It Starts

Most organizations can say how many agents they have deployed. Almost none can say how those agents talk to one another.

Two agents can be correct in isolation and still work against each other when they touch the same task, because nothing was built for them to check in first.

Communication is not a feature you add once agents are live. It is the difference between a fleet and a system.

Debt #2: Context Keeps Them Honest

An agent only knows what it has been handed, and that information ages faster than teams remember to refresh it.

Without a shared way for context to move between agents in real time, every agent works from its own version of the truth. Decisions still get made using data that stopped being current, and nobody notices until the output no longer matches reality.

Debt #3: Identity Determines Who Is Actually Acting

As agents multiply, so does the question of what an agent is authorized to do, on whose behalf and with what standing.

Without a clear identity model, an agent’s actions blur into the organization’s actions with no clean line between them. That is not a minor gap. It is the difference between knowing exactly who or what took an action and reconstructing it later from fragments.

Debt #4: Delegation Decides Whether Agents Work As A Team

The most interesting question in agentic AI is not what a single agent can do well. It is what that agent knows to hand off, and to whom.

Without a designed delegation model, agents either hold onto work they should have escalated or drop it entirely, assuming someone else has it.

Designed deliberately, delegation turns capable individual agents into an actual team.

Debt #5: Governance Is Where Ownership Either Exists Or Does Not

Most companies can name an owner for every individual agent and still have no one accountable for how those agents behave together. That gap is where debt hides until something goes wrong across agents and everyone can say their piece worked as designed.

Ownership cannot stop at the agent level. Someone has to be responsible for the system itself, not just individual agents. Without that function, every other gap stays open by default.

The consequence shows up first as lost trust. When an agent visibly gets something wrong, people stop relying on it and route around it by hand, while the system keeps running. You end up paying for automation that has become theater because no one owns the outcome.

Our experience with a distributed agent system exposed this gap clearly. Multiple agents were contributing to a business decision across functions; one agent had customer history, another operational data and a third current business rules. Each produced a reasonable recommendation, but the system’s context was inconsistent. An outdated piece of information carried into the final decision, creating an issue no individual agent had caused on its own.

When we investigated, the issue was not any agent’s logic. It was context management and governance. Nobody had defined which source of context should be treated as authoritative, how it should move between agents or when a human needed to step in before the decision went out the door.

We addressed it by establishing a shared context model, clear ownership, freshness rules and governance checkpoints. Our learning was simple: Context cannot be managed agent by agent. It must be governed at the system level, because every agent inherits context that can ultimately shape the final decision.

Turning Debt Into An Outcome

The goal was never to avoid agent debt. It is to architect for it early, the way a serious builder designs for scale.

The biggest lever is integration, where every agent shares context and reports into a place a human can see. Close behind is ownership, without which integration is simply a more expensive blind spot.

In the enterprise AI systems we build, delegation and governance are the two debts I see underestimated most, and they go together. A delegation model with no named owner is just a set of rules nobody enforces.

That gap becomes more visible as organizations move toward multi-agent AI, where multiple agents may act across the same workflow.

Left unmanaged, agent debt is a liability. Architected for and owned on purpose, it can become a strategic advantage. The companies that succeed with AI agents will know exactly what they have taken on and who is accountable for it.

Agent debt is already building behind the scenes. Who will notice first—you or your customers?

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

The Iran war that began at the end of February sent fuel prices to multi-year if not record highs, including for diesel, gasoline and jet fuel, airlines' biggest expense after labor. Supply scares with the Strait of Hormuz disruptions lasting most of this year and high demand have sent jet fuel prices and other distilled products up more than crude oil.

Airline executives aren't expecting much relief in fuel costs anytime soon, or a drop in travel demand.

"You can't run the business on the hope that the Strait of Hormuz is going to open at a certain time," Qantas Airways CEO Vanessa Hudson told CNBC in an interview late last month. The Australian airline operates some of the world's longest flights and is planning even longer ones for next year and 2028, starting routes between London and Sydney and New York and Sydney.

"There's a backdrop of resilient demand in in this environment. We're making sure that we've got our capacity settings right," she said.

Airlines have been passing higher fuel costs along to consumers by raising fares, adding fuel surcharges or increasing their checked baggage fees. The United States' on-again-off-again truce talks with Iran since the spring have kept fuel prices volatile, but airlines are holding onto those fare gains as customers continue to book.

Carriers have also raced to add more luxurious seats on board to capitalize on demand for pricier and roomier offerings, sometimes taking down the number of standard coach seats to do so.

Slightly fewer people are flying this year compared with last, but demand overall remains strong, even with higher ticket prices. Security screenings at U.S. airports were down 1% this year through Sept. 20 compared with the same period of 2025, according to a Sept. 23 Bernstein note.

Still, the latest U.S. inflation read showed airfare rose 23.4% in August from a year earlier. Holiday travel could be up even more. As of Sept. 24, fare-tracking platform Hopper said domestic, round-trip tickets were going for $402 over Thanksgiving, up 31% from last year. U.S. Christmas fares, meanwhile, are up 23% at $452 a round-trip.

Those holiday visits are essential to many families, said Hopper economist Hayley Berg, so a lot of flyers might not like the fare but book it anyway. Berg said many customers have been locking in their Thanksgiving and Christmas flights earlier than usual.

While business travel demand has been strong in recent months and off-peak trips are no longer a savvy traveler's secret, customers' focus on those key holiday trips this year could mean there are deals in the interim, Berg said.

"To me, that signals that they're already thinking about end of year travel and not thinking about any of those filler trips, fall shoulder-season trips," she said.

Airline revenue surges, profit forecast dims

Carriers are forecasting double-digit revenue growth for the third quarter as customers spend more. That's even with slightly fewer people flying.

"I've never seen in my career, outside of recovery after maybe the pandemic or 9/11 ... seen a revenue environment in terms of year-over-year improvement," American Airlines CEO Robert Isom said at a Morgan Stanley investor conference last month. The carrier has been revamping its cabins to add more premium seats and other changes to catch up to rivals.

But despite the higher fares, Wall Street analysts have cut their profit estimates for U.S. carriers since another hike in jet fuel in the summer.

American in July said it expects to lose between 10 and 70 cents a share, on an adjusted basis, for the third quarter and cut its profit outlook for 2026.

Travelers at San Francisco International Airport in California, May 22, 2026.

David Paul Morris | Bloomberg | Getty Images

Investors and anyone looking to buy a plane ticket in the next few months will get an updated view of what's to come when airline earnings season kicks off on Friday with Delta Air Lines. The most profitable U.S. carrier, which also stands to benefit from its owned refinery, will report third-quarter results and will give a fresh outlook for the end of the year.

Wall Street's focus will be on the fourth quarter and beyond. Airlines are likely to lower their earnings outlooks for that quarter, said Savanthi Syth, airline analyst at Raymond James.

She said carriers aren't likely to make as many broad-based fare increases now, but if jet fuel remains in the $4 to $4.50 a gallon range, they're likely to cut more flights, which can end up raising fares with customers facing fewer seats to choose from.

"You're going to see more rationalization in capacity," she said.

Eyes on capacity growth

Volatile fuel costs are keeping airlines cautious. Carriers have trimmed some of their schedules to cut unprofitable or underperforming routes.

Along with fewer routes, the collapse of budget carrier Spirit Airlines in May took 1% to 2% of capacity out of the U.S. market, according to Barclays, meaning low-cost and full-service airlines alike have enjoyed more pricing power this year.

Read more CNBC airline news

• Why airfare could rise even more, but airline profits won’t
• Budget airline king Bill Franke warms to premium upgrades
• Companies spurn airlines’ cheapest business class tickets
• United’s next decision: What to do with all those Boeing 737 Max 10 seats it ordered years ago
• Record heat, crowds drive offseason boom in international travel

But should oil prices fall, investors will be watchful of whether airlines increase capacity in a big way, which could end up driving fares lower.

"We expect higher booked fares should result in similar unit revenue trends in the fourth quarter for most airlines, but elevated domestic capacity growth at American and United (+10% and +9% in current schedules) will likely be the most watched guidance by investors," Barclays airline analyst Brandon Oglenski said in a note Sept. 28. "With elevated energy prices and refining margins for jet fuel persisting, we expect capacity growth ambitions will be meaningfully curtailed by most management teams, with some potentially providing early expectations on 2027 planning assumptions." %!s()

Dr. Ding Zhao, who directs the Safe AI lab at Carnegie Mellon University, has been working on this problem for almost his entire career. Now, along with veteran startup executive Kyle Wong and machine learning engineer Simo Rachidi, he’s founded a company, Safeworld, intended to solve it.

“The safety challenge that we’re talking about is a combination of, one, really advanced generative AI probabilistic evals — how do you underwrite the risk of a probabilistic system?” Zhao says. “The second part that’s really hard is the trust part, and you need both to deploy a robot.”

Safeworld is emerging from stealth today with a seed round of more than $12 million, led by Shine Capital and a16z Speedrun, with additional investment from Box Group, Carnegie Mellon University Endowment, Innovation Endeavors and SV Angel.

“The time to build an industry safety standard is now while robots are being designed and deployed,” a16z Speedrun partner Jonathan Lai told TechCurnch. “By the time you have robots in households colliding with kids and causing safety incidents, that’s way too late.”

Safeworld’s specialty is evaluating a robotic control system in simulations that are populated with realistic human models. It’s akin to the challenge faced by companies like Tesla or Wayve, which must ensure that their vehicles respond appropriately to a variety of surprising incidents they may encounter on the road. But that will be more difficult for robots, Zhao argues, because they work in unstructured environments, and because each facility they are in will have different safety standards.

“One of the most common areas is if there is a blind corner in this particular factory,” Wong said. “What is the speed or what is the stopping distance that you need to make sure that this robot will not collide with a particular human? If a human is carrying boxes, for example, will the robot detect the human or not?”

To answer that question, Safeworld will build a digital version of that corner in a model like Genesis or MuJoCo, insert a simulation of the robot it is evaluating, driven by its real software, and then run thousands of scenarios where human models encounter the robot. That’s harder than it seems, per Zhao, because people are unpredictable.

“Tripping and falling is also a good example of something that we do a lot of testing with the simulation,” Wong said. “Otherwise, you would have to go and trip and fall for the robot, which is like a hard thing to be doing all the time.”

There are definite similarities between the platform that Safeworld is building and the tools being used internally by robot builders. The founders, however, believe that beyond their specific expertise, robot-makers will want a third-party to validate their work, if only to share information about safety cases between competitors.

“A lot of people are underestimating how hard some of these edge cases are going to be to solve,” Zhao said. “It is not the robot in the vacuum, in the demo, that we are worried about. It is the robot that is deployed at scale, with people who potentially never operated a robot before.”

Vishal Dugar, the CTO of Gritt Robotics, is developing the AI brain for robots that currently help workers install photovoltaic panels at industrial-scale solar farms, and aspires to take on more complex construction tasks. His company is partnering with Safeworld as they develop their safety simulations.

“The difficulty with most of our systems is it’s very hard to formally prove it by doing some math, writing some equations, and saying yeah, the system is verified to be safe,” Dugar says. “It necessarily has to be done empirically.”

His robots operate alongside human workers, and ensuring that a robotic arm doesn’t hit them is obviously top of mind. To verify that in practice requires considering all kinds of potential scenarios.

“Humans have many kinds of appearances,” Dugar points out. “Their bodies can be in different configurations. They could be kneeling, standing. They could be tripping and falling potentially. They could be crouching. They could be running. You have to respond to all these behaviors that humans could potentially exhibit on these sites, along with the variety of variations in human appearance, you know, clothes, size, shape, height, skin color, everything else.”

It’s still early days for both Safeworld and generative AI in robotics, and the company is still figuring out the best model for its product — a platform for external users, or a services-based approach? — but the team is confident they are taking on the right problem.

“We’ll probably be the first profitable company in this field,” Zaho says. “Because if anyone wants to deploy, they need to pay us to handle the situation.”

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence. %!s()

Auroch Digital's new FPS Warhammer 40,000: Boltgun 2 launches next week, and I've been chatting to the game's principal designer Matt Bone about some of the more "creative" environments in store. It's a sturdy rule of thumb that all the best ideas in gamedev are the ones that make your colleagues want to fill your trousers with sea urchins. Bone certainly appears to have achieved this with the map comfortingly known as Vicious Circles.

"One of the things we wanted to do with Boltgun 2 is add more variety," he told me over email. "Part of that is more diverse and exotic biomes, and another part is introducing novel mechanics. The level 'Vicious Circles' is a great example of this, and definitely one of the more challenging levels to make. The informal name the team uses for the level is 'Puzzle Box', as it's made up of a series of connected, rotating 'ring' sections, as well as a giant, slowly spinning cog that acts as the finale of the level. The player moves one of the rings at different points in the levels, unlocking different areas so that they can progress (or explore for secrets).

"These moving sections are covered in enemies and pickups, which is part of the reason coders will never forgive me for this level," Bone went on. "Getting enemies to reliably navigate and fight the player on moving, interconnected platforms is a bit of a headache, as you can imagine. And spinning a gigantic cog covered in hundreds of meshes is quite the performance challenge."

Not content with alienating the coders, Bone has also managed to earn the eternal enmity of the art team. The Vicious Circles level is set in a spaceship belonging to Boltgun 2's primary antagonist Vashtorr, a demigod who – as is often the pastime of Chaos denizens – really enjoys building machines out of body parts and vice versa. "We really leaned into this 'fleshmetal' aspect for the environment, which looks incredible and appropriately disturbing," Bone went on. "But the reason artists now also hate me is that all of this intricate level art has to work in multiple configurations, not just one. A wriggling fleshy tendril may look perfect in the original level layout, but has to then avoid clipping through walls in the level's various rotations."

"Whilst it was certainly a challenge, I absolutely think it was worth it," he concluded, with the resolute good cheer of a man reconciled to his own damnation. "Hopefully it'll be a memorable level for players!"

Speaking as one such feckless player, I feel no sympathy for Auroch Digital's rank-and-file. I've just been reading the papers, and according to some columnist or other, you've all been 'corrupting the children' - being forced to make a massive Merry-go-Round out of gall bladders is no more than you deserve. Also, I really like videogame levels that are ginormous machines, such as the Clockwork Mansion in Dishonored 2, or Anoat City (kind of) in Star Wars: Dark Forces, or Into The Abyss from Titanfall 2. Any other favourite or especially loathed examples to share?

Read more about Boltgun 2 on Steam. Here is what Caelyn thought of the first: "A fast and joyously entertaining rampage through a spot-on recreation of the Warhammer 40,000 setting, Boltgun is the FPS that 40k fans have dreamed of since the first time they inexpertly splodged too-thick paint all over a Space Marine. Pacing issues can’t prevent it from being a bloody good time." %!s()

The midterms are set to take place on Nov. 3 and could have a large impact on U.S. politics, giving Democrats the chance to take back control of Congress and curb Trump’s agenda. The Hands Off Our Vote rally on Sunday was meant to mobilize voters and also featured speeches from Robert De Niro and Mo Amer.

Taking the podium at the event, Ruffalo began: “This is a big, big deal, this election. There’s a lot hanging in the balance. And we all know that we’re going to kick their asses — but we all know that they’re going to steal it. They’re going to do everything in their power, and it isn’t going to be over the night we all win.”

The “Avengers” and “Spotlight” actor then pointed out the issue of cybersecurity in voting, calling out Elon Musk in the process. “We need to be on guard for what we may not see coming: Elon,” he said. “That’s the cybersecurity of these elections. And while I’m not an expert, I’ve been talking to experts. And the one thing I’ve heard repeatedly is that there is a vulnerability of the votes when they’re transferred from the polling places to the county headquarters and then to the states.”

In order to counter this, Ruffalo encouraged voters to join the organization Free Speech for People in challenging insecure voting systems to make sure election results are as accurate as possible and “safe from any domestic or foreign interference,” as he put it.

Ruffalo then turned his attention to Trump, saying: “We gotta get this dumbass, fool ass, tired ass, lame ass, duck ass out of office. We got to make him a lame duck! We have to stop his policies now. Another two years of this, and who knows what he’ll be doing? He’s calling to throw Democrats out of the country. And he means it! The guy means everything he says. I know we’re all tired, we are so tired, but we can’t give up now … They’re going to try and stop us from taking our country back, and we’re not going to let ’em. Trump, we’re not letting you! Hands off our fucking vote!”

At that moment, a person in the crowd yelled out, “You tell ’em, Hulk!” — referring to Ruffalo’s role in the Marvel Cinematic Universe as the giant green superhero.

Ruffalo laughed and responded, “I’m gonna Hulk out. You won’t like me when I’m angry.” He paused and added: “Hey, I’m still here! They thought I was out for the count — I’m still here, I’m still playing the Hulk.”

He concluded his speech by telling the crowd not to be afraid or intimidated — and to, of course, get out and vote.

“We can’t hide. We have to be together and we have to fight for each other and we have to be brave. Our time to be brave is coming, I’m telling you brothers and sisters,” he said. “They’re going to do everything they can to keep us in our homes. They’re going to do everything they can to make us feel small. They’re going to do everything they can to make us feel lost, to make it feel inevitable. It’s not inevitable! We’re inevitable — the truth is inevitable.”

Ruffalo is no stranger to speaking out for what he believes in. His appearance at the Hands Off Our Vote event comes just over a month after he caused controversy with his criticism of the Paramount-Warner Bros. merger, which has now been approved and is expected to close this week.

In late August, he shared a clip on Instagram of Paramount board member Safra Catz outlining “really profoundly scary technologies” that Oracle — the tech giant co-founded by Larry Ellison, father of Paramount Skydance CEO David Ellison, who is involved in the merger’s financing — used to help the Israeli military after the Oct. 7 terror attack. He also likened Larry Ellison to a “classic oligarch” and warned of the consequences if Paramount and Warner Bros. are allowed to merge, alleging that the Ellisons “are crushing workers and consolidating the wealth of the world for their own power and concentrated dominance.”

In response, Paramount accused Ruffalo of invoking “antisemitic tropes,” and he was also called out by Jewish advocacy groups. Ruffalo denied the accusation, saying in a statement that it was “appalling and fundamentally dishonest.”

After the news that the merger had closed last week, Ruffalo once again spoke out against it. “This merger will stifle creativity, weaken free speech, and cost people their jobs — it is a bad deal for this country and should never have been approved,” he wrote. “This is an incredibly disappointing outcome for the hundreds of thousands of us who stood up to block it, but it’s also not the end. This grassroots movement isn’t going to fade away and neither is our resolve.” %!s()