Aurionpro Targets In-House Coding Agents Within Next Four Months
Key highlights
- Aurionpro targets a fully in-house coding-model setup within 3-4 months.
- Management expects token costs to be removed for relevant internal coding workloads after the transition.
- Aurionpro has not added development capacity for several quarters because new tools are improving productivity.
- Data-centre work could account for about 40%-50% of TIG in the current year, versus about one-third in the prior year.
powered by IRIS

Aurionpro Solutions Limited expects to replace public closed-weight models used for internal coding with in-house models within three to four months, potentially eliminating token costs. Management said on November 02, 2026 that further infrastructure investment over the next couple of quarters should support the transition and could reduce overall AI-tool spending.
Know Aurionpro Solutions inside out
Results, valuation, peers and who's buying, in one view.
When will Aurionpro replace public coding models?
Aurionpro said it could be completely on in-house models for internal coding work within the next three to four months. Ashish Rai gave that timeframe on the November 02, 2026 earnings call while addressing the company’s use of third-party enterprise tools based on public closed-weight models.
The proposed shift is based on coding agents that Aurionpro has already built and is using internally. Management said it is developing its own infrastructure to run these internal models and expects to invest somewhat more over the next couple of quarters, rather than depend entirely on external model providers.
Aurionpro said agents tailored to its own software could be “overfitted” to its internal codebase and deliver better results than frontier models for those specific tasks. Management’s case is that a purpose-built internal agent does not need the broader capabilities of a general model when the task is regular coding within Aurionpro’s own products.
Why does Aurionpro expect AI-tool costs to decline?
Aurionpro expects total AI-tool spending to decline because lower external-model use and productivity gains could outweigh internal infrastructure costs. Management said the productivity benefit from third-party tools generally offsets the token costs it currently incurs, and it does not see significant token-cost inflation at present.
Rai said coding is a competitive market with limited scope for substantial price expansion. He added that regular coding tasks are either already near an intelligence-saturation point or could reach it within about three months, meaning further improvements from frontier models may not be necessary for those tasks.
Once in-house models are deployed, Aurionpro expects no token cost for the relevant internal coding workload. That does not make AI use cost-free, because the company will bear infrastructure costs and plans additional investment over the next couple of quarters to support its models.
The expected saving remains management’s view rather than a reported financial outcome. Aurionpro did not disclose its current token bill, its internal infrastructure budget, the number of coding agents in use, or a date when all public closed-weight coding models would cease to be used.
How does the coding-agent plan fit Aurionpro's product strategy?
Aurionpro’s coding-agent programme sits within its wider effort to create AI-native versions of its software products, particularly in banking and fintech. Management said fresh product build-outs began around the first or second quarter of the prior year and accelerated through that year across every Aurionpro product.
Management said the bulk of the fresh build-out for major products is finished or nearing completion. Aurionpro had already launched one product in trade finance and transaction banking, and management expected another product launch in the then-current quarter or early in the following quarter, subject partly to holiday timing.
Aurionpro distinguished between the model layer and opportunities above and below the model layer. At the model layer, it plans continued investment in Arya, specialised models and tabular foundation-model research because it uses those capabilities; management said its clearer monetisation focus is AI-native banking software above the model layer and data-centre engineering below it.
The Banking Software 2.0 strategy is based on management’s view that AI agents need updated enterprise architecture, tools and operating rails rather than older back-office systems. Aurionpro said it is seeing market evidence for this view and expects banking wins on its new AI-native stack, although it disclosed no contract value, client name or fixed award date.
Got a follow-up? Ask IRIS
What could alter Aurionpro's spending and staffing needs?
Aurionpro said research and development spending should increasingly shift from capitalised fresh product builds to client-funded incremental work. Management said R&D had historically represented about 8% to 10% of revenue before the company increased investment in the prior year for its AI-native product programme.
Once products are in the market, Aurionpro expects incremental development work to be funded through client projects and expensed rather than capitalised. Capitalisation has tapered but not reached zero, according to management, while regulated payments can remain an exception because the full capability must be built upfront and has a longer return cycle.
Aurionpro has not added development capacity for several quarters because management is seeing productivity gains from new tools. As fresh R&D projects reduce, management expects more employees to be available for revenue-earning banking projects instead of adding the 300 people discussed in an earlier investor question.
The capacity issue also affects the Technology Innovation Group, or TIG. Management said data-centre work was about one-third of TIG in the prior year and could account for roughly 40% to 50% in the current year, with project delivery and capacity build-out determining how quickly revenue converts.
What evidence will show whether Aurionpro's plan is working?
Aurionpro’s first operational test will be whether it confirms the three-to-four-month transition to in-house coding models without internal infrastructure costs erasing the expected reduction in AI-tool spending. The company has no separately disclosed AI-cost line, so subsequent earnings calls would need to provide progress indicators or quantify the cost effect.
Commercial evidence would also come from AI-native banking launches and contract wins. Management said it expects to report wins against the new stack “very, very soon” in India, the Middle East, Southeast Asia and, it hopes, Europe, but it did not identify customers, dates or order values.
Data-centre execution remains relevant to the wider AI strategy because Aurionpro reported billing and unbilled-revenue pressure in the first half of fiscal 2027 as large orders were being seeded. Management said the issue was under control in the third quarter and expected a very strong second half, without quantifying unbilled revenue or working-capital requirements.
Conclusion
Aurionpro’s stated three-to-four-month target positions its internal coding agents as both a cost-management measure and a productivity tool. Management expects proprietary models tailored to Aurionpro software to remove token costs for the relevant work while allowing development capacity, unchanged for several quarters, to move toward banking projects and client-funded implementation.
The next items to watch are the additional infrastructure investment over the next couple of quarters, confirmation of the in-house-model transition, and planned AI-native banking launches and wins. Management also expects data-centre work to reach about 40% to 50% of TIG in the current year, making billing progress and working-capital control important alongside the coding-agent rollout.


