AI is no longer a luxury for large technology companies; it is now embedded in how cloud platforms are designed, shipped and run, thus being available to everyone in the mid-market. But technology alone does not create a competitive advantage; leadership remains crucial to success.
AWS re:Invent 2025 shone a light on the future of technology. And the future is bright.
Having attended the event, Inflexion’s Director of Technology & Cybersecurity Mike Arshinskiy felt “how quickly AI has become foundational – not an add-on, but part of how cloud applications will be built and run.”
AI’s evolution from experimentation to integration was clear from conversations across keynotes, side sessions and the expo floor itself. But the technology’s potential will take time to translate into real(ised) value. “The tools are racing ahead,” he notes. But adoption curves lag – and as with IaaS, the barrier isn’t necessarily the tech. “It’s cultural readiness and juggling business as usual priorities,” Mike highlights.
AI-enabled platform engineering to give developers time to think
One of the strongest themes was the emergence of AI-enabled platform engineering as the next productivity transformation. Organisations have long invested in platforms to streamline delivery, but the volume of tools, processes and abstractions has added to teams’ cognitive load and this is now the biggest bottleneck. Even dedicated DevOps teams struggle to cope with challenges of maintaining stability and performance of applications.
AI agents – increasingly customisable and directly connected to internal systems, documentation and workflows – are beginning to act as genuine force multipliers. They help delivery teams self-serve, troubleshoot and resolve issues far faster, enabling platform teams to focus on higher-value work. In case of the incident resolution, AI agents help to drastically reduce time to collect information and identify culprits of the problem.
“The real win isn’t faster ticket resolution,” Mike explains. “It’s the shift in how much thinking time engineering teams get back when AI can handle the heavy lifting.”
Engineering efficiency: nuanced progress
AI-assisted development has become widespread, but adoption doesn’t equate to effectiveness. The industry has seen this pattern before: serverless, microservices and IaaS all took time to translate from enthusiastic promise into a real change in performance.
While certain tasks – such as code generation – may be faster (up to 50%), holistic progress is far less owing to a hold-up in process, governance and culture. AWS reported a 16% improvement in end-to-end cost-to-serve across the software development lifecycle. At scale this is meaningful, but it is far short of the “instant transformation” tagline sometimes associated with AI.
“AI is definitely improving engineering efficiency, but it’s not a magic wand. The organisations seeing the biggest benefit are the ones adapting operating models, not just tooling,” Mike notes.
Modernising legacy estates
For many mid-market CTOs, legacy modernisation remains the largest barrier to progress. Historically, the cost, time and risk associated with refactoring or migrating core systems has led businesses to defer decisions.
AI is beginning to change that. Improved documentation generation, dependency mapping and automated test creation are making migrations faster and far more predictable. Crucially, AI-assisted approaches are reducing risk and supporting better sequencing of transformation work.
“It’s not one-click refactoring,” Mike says, “but AI is making large-scale cloud migration, application modernisation, and technical debt reduction achievable for organisations that previously avoided them due to cost or risk.”
Word on the expo floor: AI ecosystem in full flow
Day two demonstrated just how deeply AI is now embedded across the software development lifecycle. “Almost every vendor pitched an AI-enabled product or capability, a sign of how much the barrier to integrating AI has fallen. It’s basically an AI Expo now and that’s not a bad thing,” Mike stresses.
Three areas stood out.
1. AI everywhere – and that’s annoying, but good
What once felt like over-positioning now reflects changes in fundamentals. AI is now expected in monitoring tools, ticketing systems, observability platforms and infrastructure management solutions.
2. Automated resilience and incident response
AI-powered approaches to monitoring, outage triage and customer-support automation have progressed. This is being enabled by the integration of unstructured data through MCP (model context protocol; lets the AI ‘understand’ your company’s environment and act within it) and A2A (agent to agent) protocols, allowing tools to move fluidly between logs, crash reports and support tickets. It is leading to real automated resilience, a big shift in just a year.
3. AI-driven pull-request review – with one behaviour to watch out for
AI-assisted PR (pull request) review is becoming normal, and many teams are embedding it into CI/CD pipelines. But with huge AI-generated PRs increasingly reviewed by another AI, human diligence is waning. This is to be avoided: “AI is there to help engineers move faster,” Mike emphasises, “not to remove them from the loop. Human oversight remains essential.”
