Across engineering teams, AI-assisted and agentic coding tools are helping developers generate code, build features and resolve issues faster than before. For technology leaders, this has created an attractive promise, with the potential for greater productivity, faster delivery and more value from existing teams.
But while development is accelerating, a more complex reality is emerging. This new pace of development is in turn creating more pull requests and merge requests, with features ready to be reviewed, tested and validated. However, these release cycles have not always accelerated at the same pace, meaning work is piling up downstream, accumulating as it awaits testing and quality assurance.
Often this is because, as development output has increased, testing capacity, technical enablement and ways of working have stayed the same. This is felt most keenly in situations where development and testing teams are distinct, separate disciplines – AI exacerbates the issues of this delivery model while failing to move intent, context and confidence through the software delivery lifecycle (SDLC).
Here, we explore how the software lifecycle is evolving, along with practical guidance for easing the emerging bottleneck.
Intensifying the challenge of disconnected teams
Before a tester or quality engineer can validate a change, they must first understand the intent behind it. What problem was the team trying to solve? Which scenarios are in scope? What are the edge cases?
In a tightly integrated delivery model, much of this understanding is built continuously as testers work closely with developers and product teams, able to see the feature evolve and ask questions while context is still fresh.
However, when development and testing operate as separate disciplines, with a formal handover between them, this context can be harder to maintain.
In this model, development work may be completed, moved into a testing column and picked up days or weeks later by a separate team. By then, developers may have moved on to several other features. The intent behind decisions can be harder to relay, and testers must then go back upstream to ask what was meant, which scenarios matter or how a particular behaviour should work.
AI-accelerated development intensifies this problem. If developers are producing significantly more output , but the way testing receives, understands and validates that work has not changed, the queue grows, resulting in a greater disconnect between the teams, and a lack of shared understanding.
This gap can be exacerbated when development and testing teams have different levels of technical enablement. Developers may have technical knowledge and access to engineering tooling that allows them to more quickly adopt AI-assisted methods.
However, if testing teams are less technical, they may need to wait until modern capabilities are embedded in tools. This can delay how quickly testing teams can benefit from new methods and raise concerns over whether they must become more technical to keep pace with AI-enabled development.
Two routes forward: tactical and strategic
One response to this problem is to simply increase testing capacity. As development quickens, the organisation could try to add more testers to absorb the increase. In some cases, more capacity may be needed, but it is rarely the whole answer, as this can lead to several additional tasks, such as the need to onboard new team members across complex, decoupled development and test teams. Moreover, some activities cannot simply be parallelised, and so adding more testers may not solve the issue.
If the underlying process still depends on late handovers, incomplete context and manual reconstruction of intent, adding more people simply scales an inefficient model. Instead, teams should consider how they can reduce the friction and move to a model where we see the SDLC as a process that continually builds context and knowledge.
To do this, they can consider the following tactical and strategic approaches, asking what can be improved now and how software delivery can evolve for the long-term, with AI capabilities in mind.
The tactical response: improve the current model
Many organisations cannot immediately redesign their delivery operating model. They may have separate development and testing partners, or established team structures, compliance requirements and legacy processes that make closer integration difficult.
In those environments, there are still practical steps that can reduce the bottleneck.
- Create stronger context packages
Instead of handing over a ticket with minimal information, teams can provide a richer set of context (created by AI, for example) with clear summary of intent, expected behaviour, impacted areas, edge cases, assumptions, dependencies and known risks. - Use AI to support test design
This can reduce the manual effort involved in turning requirements into test artefacts, provided the outputs are grounded in the right business context. This could also be generated as a result of the handover to better connect development and testing around a central view of validation. - Apply AI to impact analysis
By summarising what has changed, which components are affected and where similar defects have appeared before, AI can help testers prioritise their effort by identifying parts of the system that are likely to be impacted by a given change. - Revisit quality gates
If AI is being used in development, leaders need to be clear about where human review is still required, where automated checks are sufficient and where additional validation is needed.
These steps should also be supported by a clear view of team skills and capacity. As AI-enabled development accelerates, leaders will need to consider how to balance expertise across the lifecycle so testing and quality engineering can keep pace. This may mean upskilling testers, increasing the number of people within their team or investing in tools that help them apply automation and AI directly.
The strategic response: make quality part of the flow
The long-term answer is to reduce the separation between development, testing and verification, evolving the role of a tester for the AI era.
In an AI-native delivery model, quality engineers and developers partner on the intent of a change, working and developing the same context together that enables more effective downstream automation. Their focus shifts from receiving completed work and manually creating test cases towards shaping intent, identifying risk, exploring edge cases, challenging assumptions and supervising automated verification.
At Endava, this thinking is reflected in Dava.Flow, which approaches delivery as a connected, AI-enhanced flow from early signal to validated outcome. Rather than allowing context to disappear at each handover, the model emphasises the creation of reusable, agent-ready artefacts that carry intent, evidence, assumptions, decisions and governance through the lifecycle.
By connecting the work earlier, business intent is clarified before delivery accelerates, and assumptions and risks are surfaced before they become defects. Governance is built into the flow rather than applied after the fact, and human judgement remains central, while AI supports the creation, enrichment and movement of the artefacts needed to make better decisions.
Moving from accelerated code to confidence
The first wave of AI adoption in software delivery has been focused heavily on creation. However, those operating with development and testing as disconnected disciplines will struggle to create more throughput and at risk of creating new bottlenecks in the process.
Leaders must consider how they restructure the SDLC around context and knowledge that can facilitate the translation into code delivered confidently. To do this, teams need to preserve intent from the earliest stages, use AI to support verification and decision-making and ensure teams have the governance structures needed to move quickly without creating unnecessary risk.
This is the principle behind approaches like Dava.Flow, connecting the journey from signal to outcome so that development, testing and delivery are not separate hand-offs, but part of one governed, AI-enabled system.
To learn how to improve quality engineering practices to build more connected, AI-enabled delivery models, or to assess and unblock bottlenecks in your SDLC, get in touch with our experts. Or, discover more about Dava.Flow and our AI-native delivery methodology.