The rapid integration of artificial intelligence across corporate technology stacks has shifted from a theoretical frontier to a measurable operational reality, particularly within the fast-paced financial technology sector. Over the past twelve months, London-based fintech Equals—formerly known as Equals Money—has executed one of the most aggressive internal AI transformations in the industry. According to James Simcox, Chief Operations and Product Officer at Equals, the company’s AI adoption curve has skyrocketed from a baseline of absolute zero to a staggering 92% of its internal codebase now being written by artificial intelligence. This exponential acceleration underscores a broader industry trend where financial institutions are moving beyond experimental proofs of concept to hard-coded operational deployment.
However, Equals’ trajectory over the past year has not been without its strategic recalibrations. Speaking at Okta’s annual user conference in Las Vegas, Simcox detailed an evolutionary journey that mirrors the growing pains of the broader enterprise software market: an initial phase of zero adoption, followed by a rush of "AI everywhere" enthusiasm, culminating in a mature, disciplined approach focused on deploying AI strictly where it adds tangible economic and operational value. For a regulated financial institution responsible for collecting, holding, converting, and transmitting corporate funds—while simultaneously issuing wallets, corporate cards, and payment solutions to end-users—managing this technological leap requires a delicate balance between automation, security, and human oversight.
The Evolution of Equals and its Technology Stack
To understand the scale of Equals’ AI integration, one must examine the foundational architecture of the firm. Operating in a highly complex and heavily regulated space, Equals specializes in cross-border payments and treasury management solutions for businesses. Over the past four years, the company has systematically modernized its technological footprint, entirely rebuilding its infrastructure within an AWS-hosted environment. Because most of Equals’ proprietary technology is built internally, the firm maintains granular control over its systems, providing an ideal sandbox for rapid software development and automated coding pipelines.
When generative coding tools matured, Equals capitalized on its internal architecture immediately. The result was a dramatic leap in developer productivity, culminating in the current metric where over 90% of all software code is machine-generated. Yet, as Simcox noted during his panel discussions in Las Vegas, leadership quickly realized that blindly applying generative AI and autonomous agents to every business process was neither cost-effective nor strategically sound.
"So we definitely have been through that journey of no AI, AI everywhere, and then back down to AI for tasks that actually make sense, right?" Simcox remarked, capturing the pragmatic pivot that many financial technology executives are currently undergoing.
From Internal Control Panels to Multi-Agent "Courtrooms"
In the highly regulated financial services sector, the margin for error is razor-thin. Providing incorrect transaction data, misinterpreting compliance rules, or mishandling customer complaints can result in catastrophic regulatory penalties, severe reputational damage, and immediate financial loss. Consequently, Equals made a strategic decision to anchor its initial AI deployments strictly within internal processes—environments characterized by strict boundaries, high visibility, and mandatory human review loops.
One of the most innovative internal applications deployed by Equals is a multi-agent framework colloquially referred to by the engineering team as the "courtroom" model. Designed to streamline the complex resolution of customer complaints, this architecture utilizes competing autonomous agents to stress-test grievance assessments. When a customer lodges a formal complaint, one AI agent is tasked with building the strongest possible argument in favor of the customer, while a contrasting agent formulates the counter-argument based on internal policy, regulatory frameworks, and historical transaction data. A third neutral "assessing" agent evaluates both arguments, synthesizes the findings, and proposes a fair resolution path.
This system addresses a persistent challenge in financial operations: while customer service rules are formally defined, human assessments of subjective complaints often vary wildly depending on the agent handling the case. By leveraging competing agents, Equals believes it can achieve fairer, more consistent outcomes while drastically accelerating case-handling times. Crucially, a human supervisor remains embedded in the loop to review and sign off on the final determination. This internal deployment served as a controlled testing ground, allowing the firm to build institutional confidence before exposing machine-learning models to direct customer interactions.
Balancing Automation with the Human Touch in Customer Journeys
As confidence in its internal AI infrastructure matured, Equals cautiously expanded agentic workflows into customer-facing operations. The firm introduced a specialized voice agent designed to handle inbound client calls and assist with foreign exchange (FX) transaction bookings. The primary objective was to expand operational availability, offering seamless service to corporate clients operating outside traditional banking hours.
Despite this technological expansion, Equals has deliberately avoided completely eliminating human customer service representatives—a pitfall that many consumer-facing technology companies have stumbled into in recent years. For Equals, personalized human access remains a core competitive differentiator in a crowded fintech marketplace.
"Why would you choose us over another fintech? It’s because you can speak to us," Simcox emphasized during the Las Vegas conference.
This philosophy dictated the precise sequencing of their AI rollout. Equals consciously refrained from launching a generic customer-facing AI chatbot as its initial entry point into artificial intelligence. Leadership recognized that deploying an impenetrable chatbot wall sends an immediate, unintended signal to corporate clients that the institution is retreating from human engagement. By ensuring that customers are explicitly informed when they are interacting with an AI agent—and retaining a frictionless pathway to escalate to a human specialist—Equals preserves its high-touch value proposition.
Furthermore, Equals has adopted a utilitarian approach to customer interactions, recognizing that autonomous agents are not always the optimal tool for every task. In many instances, traditional, deterministic software automation yields the exact same operational outcome at a fraction of the computational cost.
"Sometimes putting an agent in a customer-facing journey isn’t actually that helpful. Basic automation would be far more useful. We get the same outcome, and I wouldn’t have to spend a bunch of tokens for literally no reason," Simcox explained.
To illustrate this operational divide, Simcox pointed to the variance in transaction complexity. A routine euro-denominated payment to a standardized banking partner in France requires little to no cognitive overhead; it can be handled reliably via basic rule-based automation. Conversely, a high-friction, legally complex payment to a jurisdiction like Sierra Leone—characterized by unpredictable local banking conditions and opaque regulatory bottlenecks—demands human investigation, contextual reasoning, and tailored alternative routing. Autonomous agents occupy the middle ground, efficiently managing structured, multi-step challenges without consuming excessive resources.
Identity, Governance, and the Strategic Partnership with Okta
The aggressive deployment of autonomous software agents fundamentally transforms how enterprises must approach digital security, access control, and identity management. As organizations transition from static software applications to autonomous agents capable of independently executing business logic, the traditional security perimeters established by corporate firewalls begin to blur. This reality placed Equals’ long-standing technical relationship with identity security provider Okta into sharp strategic focus.
During Okta’s user convention, executives including President of Products Ric Smith and CEO Todd McKinnon highlighted the company’s expanding suite of enterprise AI controls. These advancements are specifically engineered to help organizations discover hidden shadow AI agents, govern complex machine-to-machine connections, and instantly sever access privileges when an anomaly or security breach is detected. For a financial institution moving millions of dollars in corporate capital daily, these identity and access control frameworks carry existential implications.
Long before Okta’s recent feature rollouts, Equals had already laid the foundational architecture required to secure autonomous workflows. During technical Q&A sessions at the event, Simcox outlined how Equals structurally decoupled user and system authentication from core business applications. By establishing a dedicated, centralized team responsible solely for managing the identity layer, individual applications operate strictly within the strict permissions assigned to them.
At Equals, autonomous agents are treated with the same rigorous governance framework applied to human employees, yet with vastly tighter scopes of operation. While a human staff member might traditionally be granted a broad operational role spanning multiple administrative functions, an AI agent is provisioned with hyper-granular permissions tailored to a single, isolated task. For example, an agent programmed exclusively to update a client’s email address is granted read-and-write access solely to that specific data field, and every action is cryptographically attributed directly to that agent’s unique digital identity.
The Auditing Dilemma: Tracking Reasoning Over Actions
One of the most profound operational divergences between human employees and autonomous agents lies in the nature of auditing and compliance logging. In traditional corporate environments, compliance auditing focuses heavily on who performed an action and when it occurred. Organizations typically do not log the internal cognitive reasoning process of a human employee because human decision-making is inherently intuitive and decentralized.
With autonomous agents, however, Simcox argues that auditing must evolve to capture the underlying algorithmic rationale. Because Large Language Models (LLMs) and multi-agent systems operate probabilistically, simply logging that an agent executed a specific transaction or database modification is insufficient for forensic analysis. To effectively investigate system errors, compliance officers must be able to inspect why the agent chose a particular path, what parameters it evaluated, and how it weighted competing variables. To achieve this, Equals developed its own proprietary internal agent platform that continuously records both the actions taken and the logical reasoning steps generated during execution.
This internal platform is further governed by a strict set of enterprise-wide "AI rules"—a codified constitution that explicitly delineates what the company’s internal and external AI tooling is legally and operationally permitted to do.
The Threat of Algorithmic Scale and the Need for a "Kill-Switch"
Perhaps the most visceral concern articulated by Simcox regarding agentic enterprise adoption is the sheer velocity and scale at which automation failures can compound. While human operators are constrained by cognitive processing speeds and physical limitations, software agents operate at machine speed across distributed cloud environments.
"Because what would be a human making a mistake once, really slowly, could be an agent making that mistake 100,000 times at the same time and completely destroying your business by accident," Simcox warned.
This systemic vulnerability makes instantaneous intervention capabilities mandatory for modern financial institutions. Simcox specifically pointed to Okta’s newly announced administrative "kill-switch" functionality as an essential tool for mitigating systemic risk. The ability to instantly terminate an agent’s access tokens across every connected system the moment an anomalous behavior pattern is detected provides a vital safety net against cascading operational failures.
Managing the Economics of Compute and Token Economics
Beyond security and governance, financial institutions scaling artificial intelligence face a persistent economic challenge: managing the shifting and often unpredictable cost of computational resources. Unlike traditional software licensing models—where operational costs scale predictably based on seat licenses or fixed server infrastructure— generative AI introduces variable token-based pricing models that fluctuate based on query complexity, model size, and usage volume.
To maintain financial discipline, Equals routes all agent interactions through a centralized billing gateway that tracks token consumption in real time, while its proprietary internal platform logs performance metrics at the individual agent level. Despite these controls, navigating the economics of AI remains a daily administrative balancing act as cloud providers and model developers continuously alter their pricing structures.
"So we’re having to just monitor it at the moment, day by day, and make sure we don’t create environments where we accidentally spend huge amounts of cash on things that are not helpful," Simcox noted.
To optimize operational expenditures without sacrificing technological edge, Equals has implemented a tiered model-selection strategy. For routine engineering tasks, automated bug fixes, and structured code generation, the firm utilizes cost-effective open-weight models where computational costs are stable and predictable. Conversely, high-value, high-complexity reasoning tasks—such as navigating convoluted international cross-border payment compliance—are reserved for resource-intensive frontier models.
Broader Implications for the Fintech Sector
The rapid journey of Equals over the past twelve months serves as a compelling case study for the maturation of enterprise artificial intelligence. By transitioning from unbridled experimentation to disciplined, risk-managed deployment, the firm has demonstrated that massive efficiency gains—such as automating 92% of code generation—can be successfully harmonized with rigorous regulatory compliance and robust identity governance.
As the financial technology industry marches further into the era of autonomous agents, Equals’ operational blueprint highlights critical prerequisites for sustainable success: decoupling identity from applications, enforcing hyper-granular permissioning, maintaining comprehensive reasoning logs, and implementing instantaneous administrative circuit breakers. For institutions dealing in the movement of global capital, the future belongs not to those who deploy AI everywhere indiscriminately, but to those who govern it with precision.
