Abhishek Saikia Co-founder & CEO of KushoAI, on why AI is making software faster, but not necessarily safer
- 7 hours ago
- 4 min read
There is a strange contradiction happening inside modern software companies right now.
AI can build code faster than ever before. Entire APIs, workflows and features can now appear in minutes rather than weeks. But while engineering speed has exploded, confidence has not always kept up.
That is exactly the problem KushoAI is trying to solve.
Founded by Abhishek Saikia, the company sits at the intersection of AI engineering and automated testing, an area becoming increasingly important as businesses rush to deploy AI-generated software into real-world environments.
Speaking to Industry Leaders, Saikia says the industry is entering a phase where reliability may become just as valuable as speed.
“AI-generated code is not automatically unsafe,” he says. “The risk comes when teams treat generated code as ready for production too early.”
And according to him, that is already happening.
“Teams are shipping faster than they can validate”
Saikia says KushoAI originally emerged from a simple but growing engineering problem: testing was becoming the bottleneck.
“APIs were being shipped much faster than they were being tested,” he explains. “Someone still had to understand the API, write the test cases, turn them into scripts, debug failures and then update everything again every time the system changed.”
But the arrival of AI coding tools has dramatically increased the pressure.
“Teams can now create features, APIs and refactors much faster than before, but testing has not caught up at the same pace,” he says. “That gap eventually shows up in releases.”
It is a pattern he believes many companies are now facing — particularly those moving aggressively into AI-assisted development.
“The biggest gap today is between how quickly teams can create software and how confidently they can validate it before release.”
“We are absolutely in a speed-over-safety phase”
While AI tooling has transformed engineering productivity, Saikia believes the industry is still in what he describes as an “output-heavy phase”.
“When a new technology helps people move faster, the first instinct is to produce more,” he says. “That is natural.”
The problem, he explains, is that every increase in development speed creates more review, integration and testing pressure further down the pipeline.
“If those parts of the engineering process do not improve, teams can end up carrying more risk even as they move faster.”
And according to Saikia, many failures are not happening in the obvious places.
“The messy parts of software are where problems appear,” he says. “A dependency slows down, a permission check behaves differently, a payment flow retries at the wrong time or a downstream system returns something unexpected.”
Those issues often slip through traditional testing systems because many teams still focus heavily on ideal-case scenarios.
“Generated code often handles the intended path well,” he says. “Production systems also need protection around the paths nobody planned for.”
Why testing became the forgotten part of engineering
One of the more striking points Saikia raises is how underfunded and under-innovated testing has historically been compared to development itself.
“Development tooling has always had an easier story to tell,” he says. “A better IDE or framework gives teams an immediate sense of progress. Testing has always had a quieter role because prevented problems are harder to measure.”
In simple terms: businesses celebrate new features more than outages that never happened.
But AI may now be changing that equation.
“As software creation gets faster, testing becomes a major constraint on release confidence.”
That shift is already pushing reliability higher up the corporate agenda — particularly in industries like fintech, healthcare and enterprise SaaS.
“A broken workflow can affect payments, customer data, compliance reporting or business continuity,” Saikia says. “Risk becomes very concrete very quickly.”
“AI reliability” may become its own industry
One of the most fascinating developments, according to Saikia, is the emergence of “AI reliability” as a category in itself.
“As AI becomes part of real products and internal systems, reliability will need clearer ownership,” he explains.
Today, responsibility is often spread across engineering, QA, security and operations teams. But he believes larger companies will soon build dedicated functions focused specifically on AI quality and governance.
“Some companies may call it AI reliability. Others may call it AI quality or AI governance,” he says. “The label matters less than the ownership.”
He predicts that within a few years, reliability will become a board-level concern rather than simply a technical discussion happening inside engineering departments.
The future of AI development will be about proof — not promises
Despite the rapid growth of AI coding tools, Saikia does not believe businesses will continue accepting blind trust forever.
“The next phase will be more mature,” he says. “Teams will still use AI for speed, but they will also expect better proof around what was tested, what changed and what could fail.”
That idea sits at the core of KushoAI’s long-term vision.
Today, the company focuses heavily on automated API and workflow testing. But Saikia sees a future where platforms like KushoAI become continuous reliability systems sitting behind constantly evolving software environments.
“In the next few years, every serious AI-driven engineering team will need a layer that continuously asks: what changed, what could break and what needs to be validated now?”
And ultimately, he says, responsible AI development starts with accepting one uncomfortable truth.
“AI will make mistakes,” he says. “Responsible AI means building systems that are useful, testable, reviewable and safe to operate when things do not go as expected.”
Follow Abhishek on LinkedIn
