↓ Skip to main content

Roadmap for Development Teams to adopt AI

·1781 words·9 mins

Our Guest
#

Deepak is the CTO of a company that developes software for the Fintech industry. Like many CTOs he feels the pressure to adopt AI but, also like many CTOs, is unsure of where to start that journey, where that journeys ends and how to get there.

Synopisis
#

Abhinay Mehta and Deepak Dhayatker discussed Rapid Edition’s fixed trading platform, its lack of current AI products, and Deepak Dhayatker’s goal of building an AI roadmap to increase productivity across development teams by focusing on areas such as code reviews, quality, and automated testing to reduce technical debt.

Lena Sesardic, Stephen Creedon, and Abhinay Mehta shared successful examples of leveraging AI for strategic insights in roadmap prioritization and enhancing developer productivity by integrating AI plugins into CI/CD processes, such as at PaloAlto, where the tool breaks down feature plans and generates smaller pull requests.

Deepak Dhayatker stressed the paramount security concerns in the fintech space, requiring a comprehensive AI policy to prevent code leaks, and requested a blueprint for incorporating Claude Code to optimize the development team and a detailed AI roadmap covering the entire SDLC.

Full Transcript:
#

Introduction and Professional Background Abhinay Mehta began by correcting an earlier statement about a CTO, clarifying that the CTO is from a software vendor for banks, still within the fintech industry, but not directly a bank’s CTO. Deepak Dhayatker then provided their background, explaining that they work as a CTO for a company called Rapid Edition, which is a software vendor for banks and has about 60 employees, with 50 in IT. Deepak Dhayatker also mentioned they had 18 to 19 years of prior experience in front office IT at institutions like Dresna Bank, Commerce Bank, and BNP Paraba .

  • Rapid Edition Product and Roadmaps Rapid Edition’s product is a fixed trading platform, where components are sold to build a whole trading platform . Deepak Dhayatker is responsible for various roadmaps, including a project roadmap for professional services, a technical roadmap largely focused on resolving tech debt, and a desired future AI roadmap. They noted that the company currently lacks AI products, with only some developers informally using AI for coding and documentation .
  • Future AI Consultancy Plans Deepak Dhayatker expressed a long-term goal of establishing a consultancy focused on Agile and AI to enhance the efficiency of software development teams and companies. Their immediate interest is in building an AI roadmap to increase productivity across their diverse development teams (web, Java, C, DevOps, and FPGA) at Rapid Edition . Abhinay Mehta also noted that Deepak Dhayatker had been an “award nominated CTO” in the fintech industry .
  • Desired Outcomes of AI Initiative When asked how Rapid Edition would change in 6 to 12 months with a successful AI initiative, Deepak Dhayatker indicated that team productivity would increase, leading to a better product . Key improvements would include more stringent code reviews, better code quality, automated testing, and higher staff motivation from using tools that simplify cumbersome tasks like debugging production issues by dealing with log files and database queries .
  • AI, Technical Debt, and Testing Deepak Dhayatker confirmed that AI is expected to help reduce technical debt, which currently contributes to senior developers spending too much time debugging. Regarding testing, they reported good coverage for unit testing in new code, a strong suite of integration tests, and one keen developer who has implemented some AI-generated automated tests on the web team . Overall code coverage in SonarQube remains low at around 45% .
  • Software Development Life Cycle (SDLC) and CI/CD Abhinay Mehta inquired about the standard development processes across the different teams and tech stacks . Deepak Dhayatker explained that all teams generally follow an Agile-based pattern with a workflow in Jira, although not all Agile ceremonies are performed, and different teams have distinct workflows. They clarified that while all Continuous Integration/Continuous Delivery (CI/CD) pipelines use GitLab and GitLab runners, each pipeline is distinct for different projects .
  • Product Feature Determination and AI Value Deepak Dhayatker confirmed that product ownership is currently shared between themself and a senior QA person, as the company has struggled to hire a senior product manager. Feature decisions are made through a forum of team leads based on requests from sales, existing, and new clients, rated on a Confluence page . Deepak Dhayatker felt that using AI to help select which features to add to the product roadmap was not a top priority due to the small size of the relevant stage .
  • Leveraging AI for Strategic Product Insights Lena Sesardic suggested an opportunity to leverage AI by analyzing sales or client call transcripts for strategic insights to help prioritize the roadmap . Lena Sesardic noted this approach could pull deeper, rich insights that manual review might miss, referencing a similar successful implementation with a founder . Deepak Dhayatker inquired if this was an area the group could assist with .
  • AI for Developer Productivity at PaloAlto Abhinay Mehta shared experience from a project at PaloAlto focused on enhancing developer productivity by modifying CI/CD processes and developing tools to incorporate AI . They described moving from using Cursor to Claude Code, building PaloAlto-specific plugins for their development practices and SDLC processes, which developers install and use for various tasks .
  • Example of AI Plug-in for Feature Planning and Implementation Abhinay Mehta detailed a “planning plug-in” that fetches Jira ticket information, creates a plan for the feature, and then implementation plug-ins for different tech stacks (TypeScript, Python) break the plan into smaller chunks . This process addresses the verbosity of LLMs and generates smaller Pull Requests (PRs). Abhinay Mehta highlighted the utility of a Cursor bot reviewer in their GitHub CI/CD images, which automatically checks PRs and has spotted bugs missed by manual testing .
  • PR Size and AI-Generated Code Deepak Dhayatker expressed significant interest in seeing a demo of the Claude Code implementation . Stephen Creedon asked about the size of the “smaller chunks” of code generated, to which Abhinay Mehta responded that without splitting, it could generate 5,000–10,000 lines of code in one unreviewable PR. With splitting, the process involves developers curating and modifying the rational steps generated by AI, which are then implemented into separate PRs or commits, although there are currently no rules on PR size .
  • Maturity of AI Implementation Abhinay Mehta explained that the company’s AI journey began with Cursor AI, but they transitioned to Claude because Claude was actively building solutions and prescribing implementation methods . Deepak Dhayatker proposed writing a white paper or documentation detailing the journey and lessons learned at PaloAlto, suggesting the current discussion focused primarily on code generation and review, with other SDLC stages like testing needing attention .
  • Alternative Methods for SDLC and Tech Debt Brian Joyce introduced the BMAD method for the full SDLC, which uses a quizzing approach to clarify requirements . Brian Joyce also suggested a technique for reducing technical debt by taking semi-large code blocks and using AI to generate pseudo-code, which makes debugging easier . Abhinay Mehta confirmed that their described process works well for both greenfield and existing code, even allowing themself (not a front-end developer) to instruct the AI to complete front-end features .
  • Prioritizing Roadmap Features Lena Sesardic circled back to Deepak Dhayatker’s priority of feature selection, who emphasized that picking the wrong feature is costly for a small firm like theirs, potentially wasting six months. The core concern is maximizing monetization and using constrained resources efficiently . Deepak Dhayatker noted that the company’s professional services team, which helps clients onboard the product, accounts for a large part of the revenue .
  • Group Purpose and Approach Stephen Creedon clarified that the group consists of individuals interested in AI and does not currently operate as a company pitching for work; their main goal is market research and understanding problems . Deepak Dhayatker challenged this, noting that CTOs with current problems usually prefer working with experts in a specific niche who offer a focused solution, rather than engaging a consultancy that needs six to seven months to define and build a solution at the client’s cost . Stephen Creedon acknowledged this, reiterating that the conversations are informing the group’s collective end goal .
  • Key Pain Points and Need for an AI Roadmap Deepak Dhayatker identified two primary pain points: optimizing code creation/review using tools like Claude Code, and using tools to improve debugging by eliminating the need to manually sift through log files and correlate data from various services . Abhinay Mehta suggested that step one should be helping Deepak Dhayatker develop a comprehensive AI roadmap .
  • Security Concerns and AI Policy Due to being in the fintech space, Deepak Dhayatker stressed the paramount importance of security, specifically preventing proprietary code from leaking and protecting secrets like tokens and IP addresses. They emphasized the critical need for an AI policy that outlines acceptable AI usage, breach reporting, and recovery procedures, noting they had tried generating an initial policy with ChatGPT. Stephen Creedon likened the necessary policy to a GDPR-type framework regarding disclosure and handling .
  • Developer Concerns Regarding AI Abhinay Mehta admitted that there are mixed feelings among developers about AI, with many feeling the “boring part” of work is now left to humans, while the AI handles the “fun part” (implementation) . They confirmed that concerns about job replacement were raised during a company presentation, with senior leaders assuring staff that AI tools are meant to augment productivity and cross-skill people, not replace them . The historical perspective was discussed, suggesting that technological revolutions, like the advent of Ruby on Rails, typically lead to more jobs .
  • SDLC Adjustments with AI Coding Abhinay Mehta explained that the front-end team at PaloAlto welcomed AI tools because they were overwhelmed by the backlog . Regarding the SDLC, Abhinay Mehta stated that the process by which ideas flow through to development largely remained the same, with AI being a new set of tools . Stephen Creedon suggested that AI tooling necessitates a greater emphasis on upfront planning and defining the expected outcome before coding, which is generally not an overly common practice in the software world .
  • Need for AI Blueprint and Future Engagement Deepak Dhayatker summarized the meeting as very useful and requested a written blueprint or links to articles on how to incorporate Claude Code to optimize the development team . They reiterated the desire for a comprehensive AI roadmap and policy, covering the entire SDLC (documentation, testing, security scanning, professional services) to move from current minimal AI use to a fully safe and governed environment . Stephen Creedon offered to discuss and put together some ideas about next steps, and they planned to catch up later .