AI-assisted coding and low code / no code (LCNC) platforms both aim to accelerate software development, but they do so in different ways. While AI coding assistants, such as Claude, can help developers write, review and refine code more efficiently, LCNC platforms enable applications to be assembled from pre-built components and reusable functionality. Both approaches are finding practical application in capital markets firms. In this article, GreySpark Partners explores whether AI will reduce the relevance of LCNC platforms, whether AI and LCNC are likely to evolve as complementary technologies or whether the two will converge into a new generation of AI-enhanced LCNC solutions for investment banking.
By GreySpark’s Chrys Ioannou, Consultant and Rachel Lindstrom, Head of Capital Markets Intelligence practice
GreySpark’s analysis of AI adoption in capital markets typically frames the technology in successive waves. The first wave was traditional machine learning and natural language processing, which have improved pattern recognition and automation for over a decade. The second wave, was Generative AI (GenAI) and large language models (LLMs), adopted rapidly from late 2023, which extended those gains into unstructured data and natural-language interfaces. Agentic AI is emerging as a third wave and allows autonomous, multi-step task execution with humans supervising rather than executing. Financial institutions have historically chosen between three models for building capital markets systems:
- Buy Approach – the use of off-the-shelf packaged software.
- Build Approach – in-house development from the ground up.
- Buy & Build Approach – the use of accelerator frameworks such as LCNC that combine elements of both.
LCNC platforms have traditionally been built around three developer personas: the Core Developer, a highly skilled engineer working via code generation on critical functions; a Rapid Application Development (RAD) Developer, responsible for a specific, less critical area such as the user interface; and the Citizen Developer, a business user with little to no coding expertise. GreySpark’s research categorises LCNC use cases in financial services into three functional groups: Digital journeys, covering customer-facing messaging and digital helpdesk applications; business workflows, covering the automated ingestion, transformation and routing of data and documents across trading processes; and enterprise systems, covering complex, mission-critical platforms drawing on multiple data sources.
The question posed by this article (whether AI, LCNC or a hybrid of the two will dominate) is, in effect, a new iteration of this same debate:
- New Build Approach – GenAI alone most closely resembles a pure ‘build’ approach.
- New Buy & Build Approach – Agentic AI integrated into a LCNC platform, with the accumulated domain expertise that is embedded within, extends the ‘buy & build’ model into a more autonomous, intelligent version of itself.
Benefits of the New Build Approach
The primary reason financial institutions build their own software, rather than buy an off-the-shelf product, is differentiation and customisation to meet the needs of their unique set up. Before AI-assisted coding, significant time and expensive resources were needed to create the thousands of lines of code required to build a solution from the ground up. Today, GenAI tools can facilitate application development, and they can do so far more quickly than is humanly possible. Incorporated into an Integrated Developer Environment (IDE), a GenAI can suggest context-aware code or complete the code where needed. Tools that have been shown to be popular with software engineers include Github Copilot, Claude Code, Cursor, OpenAI Codex and Gemini CLI. Github Copilot was found to be the most popular tool for companies with over 10,000 employees.
Beyond acceleration, GenAI-assisted coding shortens development cycles enough that budget once spent on commodity build work can be redirected towards genuine differentiation. Routine tasks such as code generation, test-writing and documentation, previously measured in days, can now be completed in hours, freeing senior engineers to focus on architecture and design problems that require human judgement. Citi claims to have saved 100,000 developer hours a week by automating code reviews.
Legacy modernisation is arguably as valuable a use case for investment banks as new development. Most banks sit on decades of undocumented legacy systems, and GenAI’s ability to read and explain old code is proving as useful as its ability to write new code. Morgan Stanley’s Dev. GenAI has saved developers more than 280,000 hours that would otherwise have been spent deciphering outdated code.
There are risk and compliance benefits too. Automated code review catches security vulnerabilities and coding-standard violations more consistently than manual review, which matters in a regulated environment where a bug in trading or settlement code carries real financial and regulatory consequences. GenAI’s capacity to produce clear, consistent documentation also improves audit readiness, a perennial challenge for banks under regulatory scrutiny.
GenAI tools can help retain institutional knowledge that would otherwise leave with departing engineers, since a model can be pointed at a legacy codebase to explain not just what it does, but why it was built that way. This also lowers the barrier for
less experienced developers to work productively on complex, proprietary systems sooner, shortening onboarding time.
Finally, adoption is becoming a competitive necessity rather than simply an opportunity. As copilots are now standard tooling at major banks, firms that do not adopt GenAI risk a widening productivity gap relative to peers. Building institutional fluency with GenAI coding tools now should also leave banks better placed to adopt the next wave, Agentic AI, as it matures.
The New Buy & Build Approach
The success that LCNC has seen in financial services has not been driven by a single standout application. Rather, LCNC platforms have been successful because they can be applied across countless business processes. Traditionally, LCNC frameworks have given investment banks a best-of-both route to building enterprise solutions, sitting between the rigidity of off-the-shelf software and the cost and delay of building an entire technology stack from scratch. Rather than committing to unfavourable vendor contracts, slow upgrade cycles and limited flexibility or absorbing the cost and resource strain of a lengthy in-house build, banks have used LCNC platforms to rapidly develop and deploy fully functional enterprise systems while retaining the ability to expand, remove or customise components as requirements evolve. Over time, LCNC platforms have also proved their resilience at scale, giving banks confidence that these frameworks can support mission-critical, end-to-end operations rather than just narrow point solutions. Consequently, LCNC platforms are helping banks to build genuinely differentiating, in-house solutions without becoming dependent on a single vendor’s roadmap or draining engineering resources away from core revenue-generating work.
The main risks associated with LCNC adoption are well established. First, non-specialist developers may lack the expertise to build robust, durable applications and second, functionality can be limited relative to bespoke development. These risks can be overcome when a LCNC framework is utilised with an Agentic AI overlay.
The first risk is mitigated by the solution’s ability to reduce the impact of user-coding knowledge distinctions. Citizen Developers have access to capabilities that previously required a Core Developer’s involvement, while giving Core Developers an assistant capable of independently handling the more repetitive elements of platform-building. The incorporation of GenAI codebuilding capabilities into the Agentic AI-enhanced LCNC framework can reduce the functionality limitations of a purely LCNC platform, which reduces the second key risk.
Instead of drag-and-drop development, Agentic AI reads the Jira files or conversational prompts to build applications from natural language. Instead of needing pre-defined rules for workflow automation, the Agentic AI-infused LCNC solution decides the next action and dynamically improves its decisions following feedback. Monitoring and governing AI and its decisions within user role limits should be part of the LCNC package. Ultimately, Agentic AI-enhanced LCNC platforms enable business users to build more sophisticated solutions and they empower both technical and non-technical users.
Agentic AI is not always a useful addition to a LCNC solution.
The creation of digital journeys and business workflows, for instance, as they are narrower in scope, may be best created as a build project using GenAI from the ground up. Enterprise Systems, however, require the orchestration, governance and resilience that mature LCNC platforms have spent years building. Consequently, the accumulated industry knowledge of a LCNC platform enhanced with Agentic AI is more likely to be the optimal solution than to build from scratch.
Figure 1: Differentiating LCNC and LCNC with Integrated Agentic AI Capabilities
Source: GreySpark analysis
Agentic AI helps LCNC solutions take actions autonomously, adhering to rules and strict user access profiles. It extends LCNC platforms from a development framework and toolset into an intelligent orchestration and automation platform, capable
of supporting end-to-end business operations, as illustrated in Figure 1.
Layering Agentic AI onto LCNC platforms does introduce a further risk, namely explainability. Regulators are increasingly expecting financial institutions to demonstrate not only what action an AI agent took, but why it took that action. Any LCNC platform that incorporates Agentic AI will therefore need to build in audit trails and decision logging as a core capability, not an afterthought, particularly for enterprise system use cases that touch regulated business processes.
Convergence Predicted to Win in Most Enterprise Scenarios
Like most things, the reasons to use a LCNC solution to create a platform are still applicable. The most realistic scenario of the three posed at the start of this article is that Agentic AI is likely to be incorporated into most LCNC platforms as an enhancement. Building an enterprise system using LCNC software, without Agentic AI, means the developers are unable to leverage the evolving capabilities to which others have access. The ‘build’ approach is likely to survive to create point solutions from scratch, however, coding without the use of GenAI is likely to be a thing of the past.
GreySpark believes that, as with most other vendor products, AI will not bring most mature LCNC products to a close; rather, it will enhance their capabilities. Vendors that can demonstrate a credible AI roadmap, rather than a bolted-on chatbot, are likely to separate themselves from the rest of the field over the next 12 to 24 months. Finally, financial institutions should not underestimate the wealth of experience that experts, from both vendor firms and from the bank’s business and IT teams, bring to the process of development.
