AI Engineering4 min
Tracking what comes back
An agent reading a job seeker’s inbox has access to the most sensitive thing they own during a search. Design for that first.
Read moreCareer guides thatrespect your time
Updated
Page 8 of 11 · 205 guides in total
AI Engineering5 min
A cold email agent, built carefully
Generating the message is the easy tenth of this. The other nine tenths is sending infrastructure and the rules you enforce on yourself.
Read moreAI Engineering4 min
Following up without becoming a nuisance
One well-timed follow-up helps. The second rarely does, and the third actively costs the candidate something.
Read moreAI Engineering5 min
Outreach on a platform you do not control
The technical question is how to automate this. The answer is that you largely should not, and the interesting design is what you build instead.
Read moreAI Engineering5 min
Outreach that is not spam
The technology for automated outreach is trivial. Everything that determines whether it works — or gets your domain blocked — is policy.
Read moreAI Engineering4 min
Building on Claude models
The features that change the economics of a job agent are prompt caching and a tool protocol you do not have to reinvent.
Read moreAI Engineering5 min
A job agent on .NET
.NET is unfashionable for agent work and genuinely well suited to the parts that are hard: long-running processes, resilience and operability.
Read moreAI Engineering4 min
Building on Gemini models
A very large context window is genuinely useful here and also a trap: it makes the lazy design possible for longer than it should be.
Read moreAI Engineering5 min
A job agent inside Next.js
Next.js is an excellent front door for an agent and a poor place to run one. Knowing where the line falls saves a rewrite.
Read moreAI Engineering5 min
A job agent in Node.js
If your product is already a TypeScript web app, building the agent in Node removes a whole category of duplication.
Read moreAI Engineering4 min
Building on OpenAI models
The features that matter for an agent are the boring ones: reliable tool calling and guaranteed output shapes.
Read moreAI Engineering4 min
A job agent in Python
Python is the default for agent work for good reasons, and the interesting decisions are about structure and concurrency rather than libraries.
Read moreAI Engineering5 min
From search to submission
Each stage of the loop fails in its own way, and the failures compound quietly — which is why the last stage is a human.
Read moreAI Engineering5 min
Combining a model with a browser
The design question is not whether to use a model. It is which of the steps genuinely need one.
Read moreAI Engineering5 min
Searching many sources at once
Fetching from five sources is easy. Making five inconsistent, overlapping, partly stale feeds into one trustworthy list is the work.
Read moreAI Engineering5 min
Finding jobs from a CV
A CV describes where someone has been. Searching from it alone finds more of the same, which is not always what they want.
Read moreAI Engineering5 min
Where a model belongs in an aggregator
Running a model over every listing is the obvious design and the wrong one. Most listings do not need it.
Read moreResumes & ATS5 min
A tailoring agent, end to end
An agent differs from a generator in that it decides what to change before changing anything — and can conclude that the answer is nothing.
Read moreAI & Careers5 min
Three overlapping job titles
Employers use these three titles interchangeably, which makes reading the responsibilities far more useful than reading the title.
Read moreAI Engineering5 min
Agents on both sides of the market
When both sides have agents, the platform is no longer a neutral venue — it is running both negotiators.
Read morePut the advice to work
Reading about ATS screening helps. Running your actual resume through one helps more.
New guides added regularly