Frequently Asked Questions

How do we introduce AI coding agents like Claude Code or Cursor into an existing dev team?

Not with training slides. We embed in your team and work real tasks from your real backlog with the agentic toolbox: Claude Code, Cursor, multi-agent workflows. Prompting precision, reviewing agent-written code and knowing when to trust the agent are skills learned through practice, not workshops. After 100 days, the toolbox belongs to your team.

Will agentic coding lower our code quality?

Only if your process stays where it was. When an agent produces 500 lines in five minutes, review discipline, test coverage and architecture ownership matter more, not less. We measure quality the same way we measure speed: cycle time, defect rates, throughput. Numbers, not gut feeling. Blind trust in agent output is wrong; so is blind distrust.

How do we measure whether AI actually makes the team faster?

Throughput and cycle time: how many tasks ship per week, and how long a task takes from started to in-the-customer's-hands. Not story points, not velocity: those measure estimation culture, not delivery. At Novadex we restructured the dev organization and brought cycle time down to 10 days, measured, not estimated.

What is the difference between Copilot and agentic coding?

Copilot is autocomplete: the developer still writes every line, just faster. Agentic coding means agents write entire modules, tests and infrastructure configuration while the developer defines goals, sets constraints and reviews results. The skill shifts from typing code to orchestrating agents: less syntax, more architecture.

Detailed comparison: Copilot vs. Claude Code vs. Cursor →

Our team is stuck between legacy maintenance and new requirements. Where do we start?

With what you have. No big-bang rewrite, no two-day framework training. We map your current flow, find the bottleneck and define measurable 100-day goals around it. Kanban method: start where you are, improve continuously, measure everything.

Do we need to hire AI specialists, or can our team learn this?

Your team can learn it; that is the entire point of our model. We bring senior specialists who work alongside your developers on real tasks, and knowledge transfer is built into the engagement from day one. We plan our exit before we start. When we leave, your team runs the new toolchain without us.

How do I assess the AI readiness of a portfolio company's dev org?

Tool licenses are not readiness. What counts: are agents part of the daily workflow, has code review adapted to agent-generated volume, and does the team measure delivery with flow metrics? We assess exactly this (architecture, process, AI leverage) and, unlike a pure audit, fix what we find in the same engagement.

What should technical due diligence cover in the AI era?

The classics (architecture, scalability, key-person risk) plus three new questions: how much AI leverage is in the development process, can the product be reached by AI agents (API, machine-readable, transactable) or only by humans with browsers, and how exposed is the business to AI-driven shifts in its distribution channels.

The full checklist: technical due diligence in the AI era →

Assessment reports keep ending up in drawers. What is different here?

The problem was never the diagnosis; it is that nobody stays to fix it. We do both: assess the dev org, then run a structured 100-day improvement with the same team that did the assessment. Goals are set on day 1 and measured on day 80. Assessment to action, not a report.

How quickly can an underperforming dev org improve measurably?

In 100 days: that is the design, not a slogan. Measurable goals are agreed before day one, reviewed at day 80, and the engagement ends with a planned exit and full knowledge transfer. No permanent dependency, no open-ended mandate. The portfolio company owns the result.