Do yourself a favor! – Learn AI Engineering – Build an agent that fixes one painful, boring workflow – Give it tools, context, memory and MCP – Write evals before you write prompts – Deploy it with a live link anyone can try – Route easy requests to small models, hard ones to reasoning models – Cap every run with token, time and cost budgets – Cache repeat queries so they cost almost nothing – Retry failed tool calls with backoff not blind loops – Validate every tool output against a strict schema – Make agent steps resumable so crashes don't restart everything – Run every tool call in a sandbox with least-privilege access – Strip secrets and PII before anything reaches the model – Treat every web page and document as possible prompt injection – Require human approval before money, emails or deletes – Log every model call, tool call and decision with traces – Score the agent's path, not just its final answer – Build a golden test set from real user failures – Block deploys in CI when eval scores drop – A/B test new models on a slice of real traffic first – Track cost per task not cost per token – Stream responses so users never stare at a spinner – Turn every thumbs-down into a new eval case – Fine-tune small models on your best production outputs – Add a fallback model for when your main provider goes down – Build a browser fallback for apps with no API – Version your prompts like code – Write a README that explains the architecture in 5 minutes – Record a 60-second demo of it working – Share the code on GitHub and add it to your resume You won't believe how many job offers you'll get. Bookmark this list. Build it all.