7 learning swaps for AI engineers who use coding agents
Keep learning to code and debug. Then move your study time from syntax drills and tutorial copying to the skills that decide whether a coding agent's work is right.
Why it matters
Coding agents write more of the code, so more of your time goes to telling them what to build and checking what they built.
OpenAI's harness engineering post and Anthropic's context engineering post both put the work in the environment, the context and the feedback loop around the agent.
These seven swaps are CloudYeti's learning priorities, not a quote from either company. The fundamentals stay.
Do this
- Pick the one swap from the list below where you are weakest.
- Spend your next week of study time on that one, not on a new prompt trick.
- Practise it on a real task: write the spec and tests first, then let the agent build.
- Check the result yourself: run the app, test the failure cases, read what changed.
- Move to the next swap when that check feels routine.
The 7 swaps, to keep as a checklist
1. Less syntax drills, more system design: where data lives and how failures spread. 2. Less prompt tricks, more context engineering: the right files, tools and constraints. 3. Less vague requests, more clear specifications: the result and the tests before the agent starts. 4. Less tutorial copying, more agent environments: tools, permissions and isolated workspaces. 5. Less looks-good reviews, more verification: run the app, test failures, check what changed. 6. Less watching every step, more reliable task loops: checkpoints, retry limits and a human to escalate to. 7. Less leaderboard chasing, more cost and product judgment: cost per successful task and value to users.
One thing to keep in mind: None of this replaces learning to code. You can only check an agent's work if you can read and debug it yourself.
The seven swaps are our synthesis. The screenshot in the video is Boris Cherny's post about plan mode, shown as an example of planning before building, not as the source of the list. Source: OpenAI: Harness engineering; Anthropic: Effective context engineering for AI agents.
One-day workshopsLearn it live, then run it yourselfLoop and harness engineering, AI agents on AWS, AI cost and ROI.See the workshops → Ask me anything about AI and cloudA paid call on your own setup: your agents, your bill, your plan.Book a call →
All tips in: Claude Code tips from the team that builds Claude Code