The Stack Is Deciding Before You Do

AI agents are moving from assistants to autonomous decision-makers — and the telecom-fintech infrastructure layer is rewiring faster than the applications on top.

# The Stack Is Deciding Before You Do

Alipay processed 120 million transactions during Chinese New Year week in January 2026 using AI Pay, a product that eliminates the checkout flow entirely. Users complete purchases through natural language commands. No buttons. No confirmation screens. The AI interprets intent, routes payment, and closes the transaction.

The interesting part: this ran on infrastructure that didn't exist eighteen months ago.

While banks debate AI adoption in committee meetings, the stack beneath them — telecom networks, payment rails, identity layers — is being rewritten by companies that control the pipes. The gap between infrastructure velocity and application-layer caution is now a strategic liability. The organizations moving fast aren't the ones with the best AI models. They're the ones that own the underlying rails and can ship atomic changes without coordinating across legacy committees.

Infrastructure moves first, applications catch up later

O2 Czechia created a dedicated AI Director role at C-suite level in early 2026. Nokia opened an R&D center in Riyadh focused exclusively on AI-powered network automation — Service Management and Orchestration (SMO), Self-Organizing Networks (SON), and AI-native operational tools. Nvidia invested $3.5 billion into MediaTek to accelerate AI capabilities in telecom infrastructure and edge computing.

These aren't digital transformation projects. They're infrastructure re-platforming efforts that change what's possible for every application riding on top.

Telcos have always controlled latency, bandwidth, and coverage. Now they're embedding intelligence into those primitives. When O2 elevates AI from an IT function to a board-level priority, it signals that network operators see their role shifting from dumb pipes to decision-making substrates. The carrier network becomes a reasoning layer, not just a transport medium.

The contrast with financial services is sharp. Rillion's 2026 AI in Finance report shows widespread AI investment across financial institutions, but also exposes significant implementation struggles and trust deficits. Banks are buying models but can't deploy them at the pace required to compete with infrastructure players who ship changes in weeks, not quarters.

Regulation as catalyst, not barrier

OutSystems launched a governed agentic AI platform for banking loan applications in early 2026. The product brings autonomous AI agents into regulated financial processes for the first time, addressing the gap between AI capability and compliance frameworks.

This is the unlock. Regulated industries haven't been slow because they lack technical capability. They've been slow because autonomy and auditability have been in tension. An agent that decides on its own is hard to explain to a regulator. OutSystems built governance tooling directly into the agent architecture, making the decision trail auditable by design.

The global digital lending market is projected to grow from $566.52 billion in 2026 to $985.03 billion by 2031 at an 11.68% CAGR. Over 90% of digital loan applications are now processed through automated AI underwriting. The volume is already there. The missing piece has been provenance — the ability to trace why an agent made a specific call.

Once governance gets solved at the infrastructure layer, everything above it accelerates. You stop asking "Can we use AI for this regulated process?" and start asking "Which processes don't need human review anymore?"

When governance gets solved at the infrastructure layer, everything above it accelerates. You stop asking whether you can use AI for regulated processes and start asking which processes don't need human review anymore.

SuperApps are shipping AI products, not AI features

Kakao upgraded ChatGPT for Kakao in January 2026 with customizable AI agent tools called Kakao Tools. The integration connects KakaoTalk Gift, KakaoMap, Melon, and external platforms like Olive Young. Users build workflows across services without leaving the chat interface.

MoonPay integrated Kamino's lending protocol — holding $1.3 billion in TVL and $1 billion in active loans — into its PayBox AI payment vault. ChatGPT and Claude users can now supply digital assets to lending markets or borrow against collateral through conversational commands.

These aren't feature additions. They're re-architecting the relationship between user intent and service fulfillment. The traditional flow — open app, navigate menu, fill form, confirm action — is being replaced by agent-mediated execution. The user states a goal. The agent orchestrates across services to fulfill it.

SuperApps have always bundled services. The shift now is that the bundle is being accessed through a reasoning layer instead of a navigation layer. The app doesn't present options for the user to choose. It interprets intent and executes. The density of what you can accomplish in a single interaction goes up by an order of magnitude.

This only works if the underlying services expose agent-friendly APIs and the infrastructure can handle orchestration at scale. Alipay's 120 million transactions in one week prove the throughput is already there.

The production engineering gap

Stampli cut product launch production hours by 68% using OpenAI's Codex and ChatGPT Work, compressing 243 hours into 77 hours for its Deep Finance product rollout. The AI tools generated code, documentation, and test cases that previously required manual engineering.

Delfi launched an AI-native balance sheet analytics and execution platform at FinovateFall 2026, targeting CFOs and treasury teams with real-time financial intelligence. The platform uses machine learning to automate balance sheet analysis that traditionally required analyst teams.

The bottleneck in finance hasn't been model accuracy. It's been the production engineering required to go from prototype to deployed system. Building the data pipelines, handling edge cases, writing the test coverage, maintaining the monitoring — that's where cycles disappear.

Stampli's numbers are specific and repeatable. A 68% reduction in production hours doesn't come from better models. It comes from better tooling around the models. The fintech companies moving fast are treating AI as a build accelerator, not just a feature. They're applying it to the engineering process itself, which compounds every subsequent product launch.

The Rillion report's finding about implementation struggles makes sense in this context. Financial institutions bought models but didn't invest in the production scaffolding required to operate them. They're trying to deploy AI into legacy build processes that weren't designed for rapid iteration. The companies winning are rebuilding the production layer first, then adding intelligence on top.

What changes when the rails are intelligent

The through-line across these stories: intelligence is moving down the stack. Telecom infrastructure, payment rails, identity layers, and orchestration platforms are embedding reasoning capability at the primitive level. This changes what applications can assume.

When Nvidia puts $3.5 billion into MediaTek for telecom AI infrastructure, it's betting that the network itself becomes a compute platform. When Nokia opens an R&D center for AI-powered network automation in Saudi Arabia, it's building tools that let carriers run autonomous operations. When O2 Czechia creates a C-suite AI Director role, it's recognizing that intelligence at the infrastructure layer is now a competitive differentiator, not a cost center.

Financial services have two paths. The first is to keep treating AI as an application-layer enhancement — chatbots, fraud detection, recommendation engines. This approach runs into the Rillion report's findings: slow adoption, trust deficits, implementation gaps. The second path is to treat AI as infrastructure — something that changes the primitives you build on top of.

OutSystems chose the second path. So did Delfi. So did Stampli. They're not adding AI features to existing products. They're rebuilding products on an AI-native foundation where automation and governance are architected together from the start.

The velocity gap between these approaches is widening. Alipay shipping an AI-native payment product that processes 120 million transactions in its first week demonstrates what's possible when you own the full stack and can move atomically. Banks coordinating across cores, channels, and compliance functions can't match that cycle time unless they fundamentally restructure how decisions get made.

Where to play

The emerging market opportunity is at the intersection of telecom infrastructure and financial services. These are regulated, capital-intensive sectors where incumbents move slowly and the infrastructure refresh cycle is measured in years. That creates an opening for operators who can ship intelligence into the pipes before applications catch up.

Nokia's Riyadh R&D center and Nvidia's MediaTek investment signal where capital is flowing. Telecom infrastructure is being re-platformed as an intelligent substrate. The operators who get there first will control the primitives that fintech applications depend on — authentication, orchestration, real-time decisioning, latency-sensitive compute.

For financial services leaders, the question is whether you're building on top of intelligent rails or trying to add intelligence on top of dumb ones. If your payment infrastructure, underwriting systems, and treasury operations assume human-in-the-loop decision flows, you're architecting for a world that's already shifting.

The companies shipping governed agentic systems, AI-native products, and production-accelerated builds aren't waiting for perfect models or complete regulatory clarity. They're solving governance and velocity in parallel, which is the only way to close the implementation gap the Rillion report documented.

The stack is deciding before you do. The operators who recognize this are rebuilding from the infrastructure up. The ones who don't will find themselves building applications on primitives that can't support the cycle time required to compete.

If you're running commercial, product, or technology strategy in telecom or fintech, the move is to map your dependencies. Identify which parts of your stack assume manual decisioning, which APIs expose agent-friendly interfaces, and where your production engineering bottlenecks live. Then decide whether you're going to own the intelligent rails or build on someone else's.