Is agentic AI undermining offshoring? Open Reply asks the question

Open Reply describes itself as an AI-first product engineering consultancy. It helps customers build digital products, by collaboratively working with clients on any part of the software development lifecycle (SDLC) from ideation, to launch and operation of the products. As such it will from time to time bump up against offshore competition.

The firm has recently commissioned a Forrester Consulting report entitled: From Code to Control: AI’s Takeover of Software Development Lifecycle to evaluate the adoption of AI in the SLDC. One of the findings of the report is that, the core rationale for offshoring – "cheaper human labor for coding/testing" – is "weakening”  as a result of the adoption of AI in the lifecycle.

No point in waiting for AI to mature

Since 2022 and the launch of gen AI tools, the technology has increasingly been used by designers and engineers to write and maintain coding output. In the last twelve months there has been an explosive growth in the capability of AI in this area which means code can be developed three to four times more quickly.

For Rhys John, Partner, Open Reply UK, his 'go to' tool is Claude Code:

It brings the focus back to the command line, pulling it into the engineering processes, keeping it simple and plugging into shell scripts. This tends to be more productive than using a copilot embedded in integrated development environments.

John has found that AI is very helpful in modernization work:

It typically takes 18-to-24 months to modernize software, but with AI you can run prototyping sessions to demonstrate how it can speed up the process. From an engineering perspective most people suddenly get it and realise it is a tool that takes away the burdensome stuff that engineers do not generally like doing. It means they can work on bigger project spaces. Engineers warm to it by using it to spot patterns and support ways that they work.

He's seemingly referring here to large or complex modernizations. Shorter projects can run in months. 

In the SDLC, John comments that while CI/CD remains very useful in coding, software engineering has not changed that much over the past decade or so. Now Gen AI is introducing new ways of doing things with its non-deterministic models so; 

You are no longer purchasing a piece of software you are planning for a continually changing technology. You are moving away from templating and moving towards flexible models. It truly will make code resilient to change. This speed of change is not going to change – it won’t slow down so there is no point waiting for the technology to mature.

AI’s takeover of the software development lifecycle

In terms of the commissioned research John explains: 

The findings show a huge appetite for adoption of AI at the top, but not yet at the developer level. The further down the chain you go, the harder it is to get value out of the tools. You have to pare things back to make sense to your own ways of working and the tools and technology you use. You need to find where AI is appropriate.

John suggests creating a 90-to-120-day roadmap to develop an SDLC for each specific project:

With engineers you need to workshop this by having a play with the tools, being hands on. You need to reset expectations. Engineers are not being swapped out for AI, it is a collaborative approach and you should start small and give the team exposure to the technology. In this way it becomes less of a mechanical threat.

You need to start at the front of the funnel. The Large Language Models (LLMs) can be part of the discovery process. You need to conduct workshops and interviews to explain how to articulate specifications in a slightly different way. Pass the bare bones of a build that the team can then run with. If you inject AI further in the process and hand it off to engineers that is more difficult to manage and will probably cause bottlenecks.

There is the wider question of the impact of AI on the outsource/insource cycle to be considered. John’s perspective is that when you outsource you are looking to reduce software development costs by using offshore or nearshore labor. But then software quality and control become a problem meaning that both bugs and technical debt begin to accrue. You also can’t test on user data offshore. He suggests that with an insourced approach in a greenfield build, agentic AI enables to the software to be developed 10-times faster than previously, although a more representative average across greenfield and brownfield builds is more like a three-times or four-times faster development shift:

In the past it has been a question of cost versus convenience, but with agentic AI, the technology can be used onshore.  Open Reply seems to be ahead of the curve with an advantage versus outsourcers in the short to medium term. Customers are finding this compelling. We know it will become a standard way of working but at the moment it is a differentiation.

My take

The research which Open Reply commissioned reports that:

93% of leaders said that their organizations plan to adopt agentic AI within their SDLC in the next two to three years as an alternative to traditional sourcing models.

It is undoubtedly true that agentic AI is changing software development and will also impact the sourcing and pricing of IT services whether they are provided onshore, nearshore or offshore. Irrespective of what shore developers work on, John’s thoughts on how to best introduce agentic AI into the SDLC are important. 

I suspect that software engineers working for organizations such as Open Reply that enable them to embrace agentic AI creatively will have an advantage over those that impose it on engineers to meet leadership technology adoption quotas. A point many software development organisations would do well to ponder.