Friday September 18, 2026
Simon L.
An AI engineer, a application engineer, an influencer partnerships squad lead, and an operations director at Hostinger stroll into a bar.
Between them they’ve got fleets of custom-built AI agents, automated workflows that substitute hours of manual work, and individual immediate libraries they’re a bit too obsessive over.
Tasks that used to obtain fractional a day now obtain minutes. Processes that needed a entire squad run in a few clicks.
They’re shipping faster, debugging smarter, and finding patterns in data that would have taken weeks to place manually.
Hence the bar. The agents grip the work, they obtain attention of the rest.
Sounds akin they’ve got it all figured out, right? Not equal close.
Agents skip steps, propose partners that have already been contacted, hallucinate particulars concerning YouTube channels, and propose code fixes that would certainly obtain downward a entire system.
One of the four is equal in a full-on beef alongside an AI delegate that’s developed an attitude.
It’s beautiful chaos. Here’s what it all really looks like.
The things they’ve built
Eylül Güleryüz is an AI engineer, which is a job heading that didn’t equal be a few years ago. It appears akin a fancy heading from the forthcoming that fair showed up early.
And it benevolent of is. She has 10 agents operating in parallel on any stated day, all on a distinct task. Her job used to be 80% her and 20% AI. That proportion has entirely flipped. Now, she’s a “manager of agents.”
“My complete betterment procedure is automated,” she says.
“I don’t equal open Jira to choice a project anymore. I have AI workflows that fig out the most impactful things to activity on, propose anywhere to start, and afterward we just… go. I verify the plan, and from there it handles everything all the way to having a drag petition prepared for me to review.”
What does all this sorcery really involve?
She calls it “harness engineering” – construction the systems that let LLMs execute tasks in the smartest and most productive way possible.
“Most of my period goes into creating the correct surroundings for the models, evaluation everything thoroughly, and figuring out new ways to shove the results further. It’s a lot additional engineering than magic words.”
Someone who’s additionally operating a fleet of agents is Karolis Varnelis. He leads influencer partnerships enlargement into new markets, which method all market is different, all spectators is different, and there’s no sole playbook that plant everywhere.
So he built one. Well, several.
“I have a guide sourcing delegate that helps discover possible influencer partners. A squad achievement inspection delegate that highlights key insights. A partner achievement inspection delegate that flags whenever achievement is dropping. A CRM audit delegate that keeps our inner CRM spotless and organized. And a satisfied norm delegate that checks whether YouTube integrations contain all the key details.”
Five agents, all alongside a particular job. That’s person who’s moved fine beyond dabbling and rebuilt how their complete function operates.
Gabrielė Bagdonaitė is the influencer achievement and operations manager, and her specialty is making achy manual processes disappear.
Exhibit A: the team’s invoice generation flow. A twelvemonth ago, they were handling invoices from complete 30 stakeholders by hand. One to three hours a day. It was as achy as it sounds. Now it’s a few actions in an AI-based workflow.
She’s additionally built agents for data checks, gathering recaps, and partner reviews, and uses AI to hunt inner conversations for precedent.
“Instead of spending everlastingly digging through messages, I can basically ask ‘have we dealt alongside this before?’ and get to the correct environment much faster.”
Then there’s Ugnius Šmaižys. He’s a application engineer, and AI has reshaped his day fair as much.
“Debugging is likely the biggest alter for me,” he says.
“Before, I could expend hours going through Grafana/Loki logs, following association IDs, checking graphs for anomalies, and tracking downward exceptions. Now I can provision AI a lot of that environment and let it nexus the dots. Something that could obtain me a few hours to track downward can sometimes obtain fair a few minutes now.”
But Ugnius hasn’t built a fleet of agents or automated a entire pipeline. He’s built a profound adequate understanding of the systems he plant on that he knows exactly anywhere AI is helpful and anywhere it’s a liability. That environment is his genuine advantage.
“People likely think the difficult part is penning the code, but normally the harder part is understanding the scheme fine adequate to cognize what code you should write.”
Four extremely distinct approaches to the identical idea: fig out anywhere AI is useful, build about it, and keep going. Of course, it additionally argues back, skips steps, and occasionally tries to scorch the home down.
The beef (and another things that break)
“I really have a ‘situation’ alongside one of my AI agents correct now. We certainly have our illusion moments and our small beefs alongside all other.” That’s Gabrielė.
She built an AI delegate to grip partner satisfied assessment – abrasion a partner’s video, checking links and coupons, verifying the merchandise is represented correctly, and logging it all in a sheet.
Sounds straightforward until you figure the systems involved: YouTube, HubSpot, coupon tools, inner platforms, distinct spreadsheet tabs.
“There are so many variables that sometimes it fair gets confused,” she says. “It completes part of the project but misses another part, or abruptly tells me it doesn’t have approval to admission a scheme whenever I KNOW it does. So we have conversations like, ‘You do have access.’ ‘I don’t have access.’ ‘Yes, you do.’ And afterward abruptly it finds the tool and continues.”
It’s akin operating alongside person who has selective amnesia but unlimited confidence.
Is she throwing it out? No. She’s breaking the workflow into smaller pieces, restructuring the logic, and simplifying things.
“Sometimes the issue isn’t that AI can’t do something. The workflow power merely be too complex, or the logic isn’t organized in the correct way for the delegate yet.”
Eylül knows this improved than anyone. When an delegate gets confused concerning permissions or skips a step, it frequently comes downward to the harness about it, which is the context, the instructions, or the constraints.
That’s anywhere a lot of her period goes, tweaking the harness until the agents do what they’re expected to.
Karolis has his own type of an agentic beef. His guide sourcing delegate keeps suggesting partners already in their system, equal whenever told not to.
“Quite often, AI provides leads that have already been shared before, equal whenever you plainly inform it not to,” he says. “You continually need to double-check, since mistakes or incorrect data can motionless happen.”
And Ugnius has caught item additional dangerous.
“AI can battle alongside systems anywhere small particulars have a lot of consequences. I’ve had cases anywhere it correctly established the logic of a issue and suggested a fix that looked entirely reasonable, but since I knew the scheme well, I could see that one small part of the alter could interrupt item much bigger.”
The code was technically accurate in isolation. It would have been extremely incorrect in context. Good fortune getting an AI to place that one.
The material they keep for themselves
They’re all profound in AI, but all one of them has drawn a row anywhere they’ve decided: this stays alongside me.
Ugnius’s row is clear. He won’t assign the thinking concerning why item is broken.
Case in point: a recurring stability matter had been hitting his systems all few weeks. He could have patched it all period alongside AI.
But Ugnius isn’t the patching type. He dug through the configuration, the logs, what the use was really doing, and eventually traced the behavior rear far adequate to acknowledge the scheme was storing a lot of data it didn’t need.
AI can discover a fix. It can’t decide whether the fix is value finding, or whether the genuine issue is location alternatively entirely.
Ugnius can since he’s throughout multiple projects at Hostinger, not fair his own. He sees how systems connect, and how a alter in one location ripples through others.
Eylül has a akin instinct. She won’t let AI anyplace near architecture preparedness for new initiatives.
“I’ll use AI to assistance build idiosyncratic pieces, but I don’t rely it to scheme and scheme the entire item for me.”
The general scheme stays hers. She’s additionally the one who, notwithstanding having automated most of her workflow, motionless makes a item of talking immediately to customers.
“There’s item really precious concerning understanding who’s really using what you build all day,” she says.
When everything alternatively runs through agents, she’s deliberately keeping one conduit individual to remain connected to the bigger picture.
Karolis motionless evaluates YouTube channels manually. AI gives generic insights and sometimes misses the nuance.
His creator partners are developers, AI automation enthusiasts, ecommerce experts creating detailed tutorials and profound dives. Knowing what makes a fine partner requires merchandise knowledge, specialized understanding, and a trained eye no example has matched yet.
“This volition enhance complete time,” he says, “but for now, a manual eye inspect motionless plant better, as we can validate the conduit according to our own cognition and experience.”
And Gabrielė? She keeps AI out of one particular place: squad brainstorms.
“I motionless really value that individual communication and those random ideas that arrive from group bouncing thoughts off all other,” she says. “Of course, AI can associate following the gathering to recap everything and assistance us arrange the chaos. But during that imaginative part, I’m blessed to let the humans do their thing.”
Too profound to go rear now
For all the friction, they’re in too profound to go rear now. And they cognize it.
Eylül stated she’d be “devastated” if person pulled the plug on AI. Gabrielė stated going rear to doing everything manually would “feel nearly illegal.”
They akin how they activity now. The speed, the range, the fact that they can build item in an midday that would have taken a week.
They’ve established a way to do their finest work, and it shows.
If that appears akin your benevolent of chaos, we’re hiring. There’s area at the bar. Gabrielė’s delegate says there isn’t. There is.
Simon is a energetic Content Writer who loves assisting group change their imaginative ideas into thriving businesses. With extended promotion experience, he continually strives to nexus the correct communication alongside the correct audience. In his spare time, Simon enjoys lengthy runs, nurturing his chilli plants, and hiking through forests. Follow him on LinkedIn.
