Why DevOps, Infrastructure Must Evolve to Manage AI-Accelerated Development
By Brian Pereira (creed_digital)
The enterprise software industry stands at an inflection point. Software development is moving from artificial intelligence-assisted development – or vibe coding – to full-stack development.
Software developers will soon collaborate with AI agent developers to accelerate development, bringing radical changes to the software development life cycle. Tools such as Replit provide a full-stack development environment for building apps, websites and games directly from your phone or computer.
There’s now discussion about the agent development life cycle and the need to supervise or manage AI agent developers – calling for agent governance and infrastructure changes. New products, services and partnerships announced in the past few weeks support this trend.
Anthropic announced its Claude Sonnet 4.5 model with advanced agentic capabilities and a feature called Checkpoints – which enables users to save progress and instantly roll back to a previous state in Claude Code. Anthropic and IBM then declared their strategic partnership that allows the tech giant to implement Claude large language models into its integrated development environment. Sam Altman announced new software development kits including AgentKit at OpenAI DevDay 2025, dropping strong clues about how enterprise users will consume apps. Google announced the launch of Gemini Enterprise.
To manage the resulting agent sprawl, CIOs and CISOs need to supervise agents and manage agent identities.
Agent sprawl is “for real,” with clients building “hundreds of thousands” of agents, said general manager of data and AI at IBM, Rikita Gunnar on a panel at the Scaleup:AI ’25 event in New York City. “Agents now have an agent development life cycle, or ADLC, and there are certain things one should do to manage this life cycle.”
It is now imperative to build a center of excellence to hone agents to perform their tasks accurately, Gunnar said. “It’s no longer just about reskilling people.”
“Every piece of software will be agentic. An MCP server changes the interaction paradigm and will call for new skill sets,” Gunnar said.
An MCP server is a component of the Model Context Protocol, and it securely exposes tools, data and functionality to AI agents and LLMs.
Enterprises were cautious about entrusting public models and agents with intellectual property. But the partnership with Anthropic could make models more trustworthy.
“Enterprises are looking for AI they can actually trust with their code, their data and their day-to-day operations,” said chief product officer at Anthropic Mike Krieger.
From AI-Assisted to Autonomous
For decades, DevOps teams reviewed and tested code through phases termed alpha, beta, release candidate and general availability. This process is still being followed, but the accelerating pace of development demands more agile development.
Over the years, more automation has been introduced in the SDLC processes. Continuous integration/continuous development, or CI/CD, is now a key part of DevOps
CI/CD automates routine software development tasks such as merging code, building, testing and deploying, enabling teams to release code changes more frequently. To increase automation, agentic AI is being embedded into CI/CD processes.
Embedding agentic AI within the fabric of enterprise architecture enables organizations to unlock transformative agility, reduce cognitive load and accelerate innovation – without compromising trust, compliance or control – says an IBM report titled “Architecting secure enterprise AI agents with MCP.”
Developers adopted globally recognized models such as Capability Maturity Model Integration, or CMMI, and CMMI-DEV as paths to improve the software development and maintenance processes.
As the timeline of development cycles compressed from months to weeks to days, crucial aspects such as data security, privacy and intellectual property protection had to be addressed. New laws and regulations also mandated data residency and sovereignty.
AI agents, regarded as modern-day digital coders, can help in these areas. That’s a leap over using generative AI for vibe coding. AI agents are being introduced at multiple stages of the SDLC.
For instance, IBM’s Project Bob adapts to developer workflows from design to deployment. Whether developers are modernizing legacy systems or building something entirely new, Bob is designed to help coders to ship quality code faster.
Adapting Infrastructure for Agents
Enterprises must be prepared to implement radical process and infrastructure changes to successfully adopt AI agents in software delivery. AI agents must be managed by a central governance framework to enable complete visibility into agents, agent performance monitoring and security.
A key finding in a 2025 report by Google Cloud’s DevOps Research and Assessment, DORA, team titled “State of AI-Assisted Software Development” shows that AI adoption improves software delivery throughput but increases delivery instability. The report suggests that while teams are adapting for speed, their underlying systems haven’t yet evolved to adequately manage AI-accelerated development.
The report confirms that AI adoption has become nearly universal, with most survey respondents – 90 per cent – saying they use AI as part of their work, and more than 80 per cent believing it has increased their productivity. Yet 30 per cent reported little to no trust in the code generated by AI, indicating a need for critical validation skills.
The research includes more than 100 hours of qualitative data and survey responses from nearly 5,000 global technology professionals.
Google’s DevOps Research and Assessment team, or DORA, identified seven capabilities that might contribute to better outcomes for AI-assisted development teams:
- Clear and communicated AI stance;
- Healthy data ecosystem;
- AI-accessible internal data;
- Strong version control;
- Working in small batches;
- User-centric focus;
- Quality internal platforms.
