Aditya Khandelwal describes a recurring enterprise pattern: a few developers gain strong leverage from coding agents, leaders mandate broader use, low-quality output and incidents follow, and token budgets are added after costs rise. He reframes the challenge around each developer's fear of displacement and confidence using agents rather than treating adoption as a tooling switch.
Aditya Khandelwal lists warning signs of a weak setup, including constant agent babysitting, unexplained swings in model quality, excessive context consumption, long sessions requiring intervention and repeated low-quality output. He argues that engineering leadership must own the shared codebase and workflow changes because individual developers cannot resolve organization-wide conventions alone.
Aditya Khandelwal recommends treating the codebase as a progressive-disclosure systemContext engineering designs how relevant instructions, evidence, state and tool information are assembled for a model's task. that gives an agent the right context when needed. He pairs this with automated checks that detect and repair quality drift, continuous iteration as models and harnesses change, and explicit attention to fear, trust and differences in developer experience.
Aditya Khandelwal's team began with shared baseline practices, invested in one high-value skillAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work. that carried finished code through pull-request preparation, connected issue tracking and continuous integration to the repository, and added recurring code review automationCode review examines proposed source changes to assess their correctness, clarity and suitability before or around integration.. He closes with practical failure modes, including issue proliferation, disagreement, long-running agents, merge conflicts and experimental code, then answers how thin instruction indexes and code-linked runbooksRepository context is the relevant source, configuration, history, documentation, dependencies, and worktree state needed to understand a software task. support progressive disclosure.
Watch on YouTube
