Sharath Kumar S has always been a builder at heart. Now a Lead Engineer on Abnormal's Messaging Security team, he's spent his time here proving what that title actually means day to day. Before joining Abnormal, he co-founded a startup with a friend, moving through cable TV software, a quantitative trading platform, and a stock portfolio management system, learning along the way how to take an idea to a working product fast and own every layer of the stack. That instinct followed him into a last-mile logistics company in India, where thin margins meant every system had to be resilient and efficient from day one.
A Different Kind of Trust
Sharath's path to Abnormal wasn't a straight line from computer science degree to security engineer. He built products end to end: backend, frontend, cloud infrastructure, and quality assurance, while also making product calls and talking directly to customers using what he shipped.
"I have worn every hat: full stack across backend, frontend, cloud, devops, and QA, but also product manager deciding what to build, and customer support talking to the people actually using it," Sharath said.
That range shaped how he approaches Messaging Security today. The domain shifted from optimizing delivery routes to protecting Slack and Teams channels from real threats, but the underlying discipline held: build resilient systems, sweat the safe default when something fails, and don't waste effort.
He walked into Abnormal expecting something narrower than what he found: a slower, more layered company, narrow tickets handed down from above, limited room for an engineer to make real calls.
"I could not have been more wrong, and it was wonderful," he said. On Messaging Security, he wasn't handed a spec. He was trusted to shape the scope, the security posture, and the rollout of an entire product surface, and encouraged to move fast on all of it.
"That kind of trust is rare, and feeling it so early was genuinely energizing. It gave me all the speed and autonomy of a startup, but with the support and scale of a real platform behind me."
The second surprise came from how the team treats tools. There's an active, shared push to find better ways of working, and people share what they discover freely.
Meet the Abnormal Intern Force
That culture is what gave Sharath the room to build something of his own. Sharath calls his system AI Force, short for Abnormal Intern Force. It gives him a tile for each AI coding session running in parallel, so he can point every one at a different piece of work and watch them all at once.
On a recent project building an entirely new product feature, Sharath split the work across sessions instead of moving through it one thread at a time. One session drove the frontend, another handled the backend, and a status panel tracked each one for build health and review status. A desktop notification pulls him in the moment something changes state, so his attention goes exactly where it's needed, when it's needed.
"You would see me running a command center, not writing code," he explained. "A problem lands, and instead of rolling up my sleeves to grind through it front to back, I break it into independent missions and hand each one to an AI intern."
One intern reproduces what broke. Another tears the change apart looking for what else might fail. A third chases the behavior through the real system to confirm the fix holds. They all run at once, while Sharath directs rather than executes.
"My job stops being execution and becomes command. I set the objectives, brief each intern sharply, and let them work in parallel while I hold the whole picture in my head."
What surprised him most was how calm the whole thing feels. "All the waiting and the busywork get absorbed, and what is left is the only part that was ever really mine: asking the right questions and judging the answers," he said.
Doing Two Jobs at Once
Six months ago, Sharath thought of his time as a single track: either heads down building a feature, or looking after the team, reviewing pull requests and unblocking people. Doing one meant the other waited.
That tradeoff is largely gone now. Because he can direct a fleet of agents in parallel, he can carry a lead's responsibilities and an individual builder's output on the same day.
"Today I would not hesitate to take on a whole new area and keep leading at the same time, because the ceiling is no longer how many hours I can personally type," he said. "It is how clearly I can set direction and judge results, and that scales in a way raw effort never did."
That's the tension a lot of lead-level roles carry: enough hands-on ownership to stay sharp technically, and enough scope to unblock a whole team. For Sharath, AI is what made both possible without sacrificing either.
All of that showed up clearest on one project: bringing Slack into Abnormal's Messaging Security coverage.
The Latency Problem That Gets Faster as You Grow
Owning Slack Messaging Security end to end meant making calls that had no existing playbook. The hardest one: how to get a message from posted to scored with the lowest possible latency, while scaling across many customer integrations and staying inside Slack's rate limits. Low latency wants constant checking. Rate limits and scale reward the opposite.
A fully push-based design would have been fastest, but the access model Abnormal needed for multi-tenant, organization-wide visibility called for a different approach. Fixed-interval polling was the easier fallback, but it doesn't scale: the more customers you add, the slower it gets.
Sharath chose a third path: distributed polling that splits work across a pool of workers by ownership, so adding capacity shortens how often each customer gets checked. Each worker resumes from the last real activity it saw, so it never wastes budget replaying old ground and never misses a message that lands mid-cycle.
"The interesting latency problems are about designing so the system gets faster as it grows, and spending a scarce budget only where it buys you freshness," he said.
That same instinct, extend what exists rather than build something new, is what lets him move fast without reinventing the platform underneath him.
"The platform was built so that a new and unusual requirement could be absorbed by extending what exists, instead of forcing a special case off to the side. Good architecture is not just the code that works today. It is how easily the next unexpected requirement fits in."
Safe Defaults, By Design
On a security surface, the quiet failures matter most. A crash gets noticed in minutes. The costly failure is when a dependency blips and the system mistakes "I could not check" for "nothing to worry about."
Sharath built every step to earn the right to act. Before touching a customer's messages, the system checks that everything it depends on is genuinely present and healthy. If anything is missing, it doesn't proceed on partial information. It records what was missing and waits to try again.
"I would rather the system honestly do nothing than confidently do the wrong thing," he said.
What He'd Tell the Next Engineer
Asked what he'd tell someone trying to build their own version of AI Force, Sharath didn't point to the parallelism or the dashboard, he pointed to attention, and to protecting it.
"The hard part, and the whole point, is making sure the system pulls you in only when a human decision is actually needed and stays quiet otherwise," he said. "If you get that wrong, you have not built a force multiplier, you have built ten things to babysit, which is worse than one."
Design around the interrupt first, he says: what pings you, when, and why. Everything else exists to protect that attention.
For anyone joining Abnormal's engineering team, his advice comes down to owning problems instead of waiting for tickets, looking for what already exists before building something new, and actually using the AI leverage available rather than treating it as optional.
Underneath all of it, the freedom and the speed only mean something because everyone points them at making the product genuinely safer for the people using it. Ownership without that focus is just noise. Here, it rarely feels like noise at all.
Want to see more of how Abnormal engineers build? Follow along on the Abnormal Builders Substack at builders.abnormal.ai.
