Murali Swaminathan, CTO, Freshworks

| Photo Credit:

Bijoy Ghosh

As it reshapes its product development around AI, Software-as-a-service(SaaS) major Freshworks is factoring in the time taken for code to move from development to production rather than raw usage to measure AI effectiveness. Speaking to businessline, Murali Swaminathan, CTO, Freshworks said that while AI can generate huge amounts of code, processes like review, testing and deployment continue to be the bottleneck for productivity gains.“We track token usage since we spend money on it, but consuming a lot of tokens alone does not give you any efficiency gains. What we saw internally was that we were able to generate a lot of code but it got stuck in the review or testing phase where a human has to go through it,” he said. Swaminathan said that the company is optimising the entire software lifecycle by introducing AI across these phases alongside coding.‘Context layer’Beyond speeding up development, he added that AI is helping the company build a ‘context layer’ that carries information across various teams including engineering, design and support allowing them to reuse knowledge across the software development lifecycle.“We realised that simply giving people AI tools and asking them to use them led to a lot of duplication and inconsistency. The focus now is on building common skills and patterns that developers can reuse rather than everybody solving the same problem differently,” he said.Meanwhile, Dennis Woodside, CEO, Freshworks said that AI has helped accelerate Freshworks’ internal development cycles by about 30 per cent with the company redeploying productivity gains towards new projects rather than workforce reductions.As for how SaaS pricing is expected to evolve with AI tools, Swaminathan suggests that while consumption based pricing is set to pick up, customers continue to expect predictability. “Consumption pricing is one model but then the question is how do I charge for that consumption. Customers want predictable outcomes and not a running bill,” he said. He suggested based on customer preference multiple models could gain traction including outcome-based pricing where charges depend on the complexity of the task.Published on June 25, 2026