Few IT executives really feel the tempo of developments in synthetic intelligence (AI) as acutely as Manu Narayan. Some 9 months into his function as the primary chief info officer (CIO) at GitLab – the software program growth platform with over $1bn in income and greater than 2,000 workers – Narayan is tasked with turning the corporate right into a proving floor for the very applied sciences its prospects use.
“The AI house typically is altering so quickly that we’ve consistently needed to revisit our targets and issues that we wish to accomplish,” he stated in a latest interview with Laptop Weekly.
With product growth sitting with GitLab’s analysis and growth group, Narayan’s mandate is usually inner: modernising the enterprise software stack, person assist, in addition to knowledge and analytics. However as an alternative of bolting AI onto current workflows, his purpose is to rebuild operations from the bottom up.
“Once I was revisiting our AI technique just a few months in the past, the main target was not on how we introduce AI,” he stated. “The main target was to rethink the character of labor internally, leveraging AI. It’s serious about processes from first rules after which utilizing agentic AI to drive them.”
Pointing to a buyer success supervisor (CSM) for instance, Narayan famous that the aim of the function is to construct deep shopper relationships, but CSMs spend hours on administrative duties comparable to constructing quarterly enterprise evaluate slides for purchasers, transcribing notes and trying to find context throughout buyer relationship administration methods, knowledge warehouses and chat channels.
By deploying AI brokers to deal with the grunt work, GitLab is seeking to unlock its workforce to give attention to high-level technique. “We would like all of our group members to give attention to what issues most: the core function of their function,” stated Narayan. “We’re leveraging AI for duties that may assist them scale out in a extra linear manner, greater than only a 10-15% improve in productiveness.”
To handle AI deployments, GitLab has adopted a hub-and-spoke working mannequin. A central AI enterprise group handles governance, technical constructing and guardrails, whereas devoted “AI transformation house owners” embedded in particular person divisions determine time-consuming, repeatable work that’s ripe for automation.
The method has already been utilized to GitLab’s personal inner worker assist community. The corporate has constructed AI brokers to help its 120 inner assist workers throughout IT, folks operations and gross sales, serving to them immediately pull the context they want or deflect routine tickets fully.
Rejecting ‘tokenmaxxing’
As AI adoption will increase throughout the enterprise, CIOs will naturally grapple with value management and measurement. Nonetheless, Narayan is cautious of methods comparable to “tokenmaxxing”, the place builders and workers are inspired to maximise the variety of AI tokens they use.
It’s straightforward to get to 90% of an software you develop in-house. That final 10% – the role-based entry controls, auditability, immutable logging, that are stuff you want as a public firm or as an organization that offers with regulated prospects – is extremely complicated to construct Manu Narayan, GitLab
“We’ve particularly averted and don’t wish to do tokenmaxxing,” stated Narayan. “Gamification will help drive outcomes, however I feel it drives the inaccurate behaviour. We’re not in search of purely context-in, context-out because the measure of success. It’s actually arduous to know if someone’s gaming the system. Are they only sending extreme content material as a result of they don’t truly know what they’re doing?”
As a substitute of monitoring token burn, GitLab tracks each day lively utilization throughout the tech stack to make sure its workforce is constructing sustainable habits. For calculating arduous return on funding (ROI), Narayan insists on anchoring AI deployments to conventional enterprise metrics. For an AI agent helping a gross sales growth consultant, success isn’t measured by the variety of prompts generated, however by commonplace key efficiency indicators: outbound messages, conferences scheduled and gross sales pipeline conversion.
Construct vs purchase and the way forward for SaaS
“We might even see extra customized interfaces and the disaggregation of methods of interplay from methods of file,” he stated. “However the underlying governance controls in core SaaS instruments aren’t going wherever.”
Narayan additionally pointed to the hidden prices of bespoke software program growth: “It’s straightforward to get to 90% of an software you develop in-house. That final 10% – the role-based entry controls, auditability, immutable logging, that are stuff you want as a public firm or as an organization that offers with regulated prospects – is extremely complicated to construct.”
To make sure security throughout customized and provider instruments, GitLab grounds its AI governance in a strict knowledge classification commonplace. Public knowledge flows by way of self-service platforms, whereas proprietary or buyer knowledge requires deeper safety opinions earlier than interacting with massive language fashions.
Regardless of robust govt backing and funds, change administration stays a problem for Narayan. Bridging the hole between AI-forward workers and those that are slower to adapt requires a mixture of departmental centres of excellence and inner AI hackathons.
But, for a CIO, the best strain is the ticking clock.
“The factor that retains me up at evening is whether or not we’re transferring quick sufficient,” stated Narayan. “Within the AI period, our decision-making must occur in days and weeks, not months and quarters. However I nonetheless fear about whether or not we’re driving the correct initiatives which might be going to have the correct long-term ROI for us.”
