Giving every developer free rein with AI tools is like adding more planes to the sky without updating air traffic control.
“The traditional software development lifecycle was built for humans: long sprints, handoffs, ceremonie. Adding AI as an assistant speeds up individual tasks but leaves those bottlenecks in place.”
Dedicatted Chief Technology Officer Serhii Semenchenko has seen this coordination problem up close. His company works with engineering teams adopting AI across software development, which gives him a close view of what happens when new tools arrive faster than the processes around them can adapt.
“In most cases, developers have licences for Claude Code, Codex, Copilot, or Cursor and use them on their own: autocomplete, generating a function, explaining legacy code,” Semenchenko said. “That’s a good start, but it covers only one part of the life cycle.”
The biggest risk, he said, is fragmentation, especially once AI moves beyond writing code and starts helping teams decide what software should do, test whether it works, and manage it after deployment.
“There’s no shared context about the company’s architecture and standards, no common rules across teams, no AI in requirements, design, testing, or operations, and no baseline to show whether delivery actually got faster,” Semenchenko said. “Without these, the team types code faster, but it doesn’t deliver faster.”
The software around the software
Dedicatted is working within the AI-Driven Development Lifecycle, or AI-DLC. AWS introduced AI-DLC in 2025 as a new methodology for bringing AI into the software development process. Dedicatted, for its part, is a certified AI-DLC ambassador for AWS.
The framework covers inception, when teams define requirements and plans; construction, when software is built and tested; and operations, when it is deployed and managed. AI enters that process from the outset.
“The traditional software development lifecycle was built for humans: long sprints, handoffs, ceremonies,” Semenchenko explained. “Adding AI as an assistant speeds up individual tasks but leaves those bottlenecks in place.”
AI-DLC is meant to keep the work connected across the entire lifecycle. The same standards, project decisions, and technical context follow the work from one stage to the next, so that each agent is not starting from scratch or relying on its own interpretation.
“It runs on a simple loop,” Semenchenko said. “AI proposes a plan, asks clarifying questions, and executes only after a human validates the plan.”
Putting it in practice
If AI agents are machines on a factory floor, AI-DLC is the set of instructions, quality checks, handoff rules, and checkpoints that determine how the whole factory runs. In the inception phase, an AI agent can take a business goal and turn it into proposed requirements and units of work. The team then reviews those proposals, answers questions, and settles important decisions before the agent moves on.
In the construction phase, AI uses that validated context to propose architecture, code, and tests, while the team resolves technical decisions. Dedicatted adds another layer by turning company standards into instructions that agents can use consistently. It has built a library covering security, infrastructure-as-code, testing, and AWS Well-Architected practices that engineers’ agents can load regardless of which AI interface they prefer.
During mob sessions, product owners, engineers, and, if needed, security or quality assurance staff review an AI-generated plan together. The agent can raise questions about issues such as authentication, data retention, or failure handling before the team approves the plan and proceeds with work.
“One session like this often replaces days of Slack and Teams threads,” Semenchenko added.
Human oversight is built into the framework at key decision points, and Dedicatted sets particularly strict limits for AI around higher-stakes decisions.
“We deliberately limit its role in decisions with business or irreversible consequences,” he said. “That includes architectural trade-offs, security exceptions, production data changes, and anything involving regulated data. In those cases, AI proposes, and a human decides, always.”
Once those standards are defined, AI-DLC can carry them into continuous integration and continuous delivery, or CI/CD. AI agents can generate tests alongside code, while security and testing requirements can be set as blocking rules that prevent a change from moving forward until problems are resolved.
“The principle is that people define the rules once and automation enforces them on every change, which removes as much of the human factor from checks as possible,” Semenchenko added.
The capability you can’t buy
Dedicatted uses AI-DLC internally and helps clients adopt the model. The company used the framework with an aviation technology company that had been taking roughly a month to turn a new idea into a working prototype, according to Semenchenko. Setting up technical environments, writing boilerplate code, and completing security reviews consumed much of that time.
Dedicatted built a system that let AI agents work from the company’s own architecture and security standards, using development environments that closely mirrored production.
“Idea to prototype went from roughly a month to days,” Semenchenko said, adding that the shorter cycle allowed the company to test several ideas at once and abandon weak ones earlier.
Once AI-DLC moves beyond a small pilot, a company has to decide who owns engineering standards, how teams review the extra output, when security gets involved, and how success will be measured.
Dedicatted tracks metrics such as the time from idea to production, pull-request review time, the share of AI-generated changes accepted without rework, failure rates, and when security issues are caught. “Lines of code generated or percentage of code written by AI are vanity metrics,” Semenchenko added.
Putting AI tools in developers’ hands is relatively simple. Semenchenko believes that the companies that get the most from AI will be the ones that have done the harder work of adapting the development process around it.
“Everyone else will write code faster and wonder why they aren’t shipping faster,” he added. “Tools are becoming a commodity. Process and context are where the competitive advantage is.”
PRESENTED BY
Want proof this works? Read more about how an aviation tech company cut proof-of-concept time from a month to days.
Feature image courtesy Dedicatted.