AI Engineering5 min
Agents that operate a computer
Operating a screen is the most general form of automation and the most expensive. Use it where nothing else reaches, not as a default.
Read moreCareer guides thatrespect your time
Updated
Page 7 of 11 · 205 guides in total
AI Engineering5 min
Navigating an applicant tracking system
Applicant tracking systems are a small number of products dressed in thousands of skins. Detect the product and most of the difficulty disappears.
Read moreAI Engineering6 min
A browser agent that applies to jobs
A browser agent that clicks Submit is holding a loaded action. Everything in the design should be about being certain before that moment.
Read moreAI Engineering5 min
A browser agent behind MCP
Publishing browser control as MCP tools means an agent you did not write can drive a browser you are responsible for.
Read moreAI Engineering5 min
Playwright as an agent runtime
Playwright was built for tests, where the author knows the page. An agent does not, which changes which of its features matter.
Read moreAI Engineering5 min
Choosing a browser automation library
The usual comparison is written for test suites. Agents need different things, and the ranking changes once you weigh those.
Read moreResumes & ATS6 min
One career, many versions
Generating forty versions is trivial. Keeping forty versions consistent with each other, and with the person, is the actual problem.
Read moreResumes & ATS5 min
Knowing before you apply
The useful question is not whether you meet every requirement. It is which of the ones you miss would actually stop you.
Read moreResumes & ATS6 min
Per-job resume rewriting
Tailoring should change what is emphasised, not what is true. The architecture is what enforces that, not the prompt.
Read moreResumes & ATS5 min
The profile everything else depends on
Every downstream feature — matching, tailoring, auto-apply — is limited by this one object. It is usually the least designed part of the product.
Read moreResumes & ATS5 min
Building an ATS resume scorer
Most ATS scorers measure two unrelated things and add them together. Separating them is what makes the output useful.
Read moreResumes & ATS5 min
Analysing gaps in a career
A tool that detects employment gaps is one design decision away from being a tool that discriminates against carers and the recently ill.
Read moreResumes & ATS5 min
Personalising an application
Inserting the company name is not personalisation. Recruiters can tell, and a template with a merge field is worse than no letter.
Read moreResumes & ATS5 min
Finding what is missing
Half of what a gap analyser finds is not a gap at all — it is something the candidate has and never wrote down.
Read moreResumes & ATS5 min
Keyword optimisation, done responsibly
The line that decides whether this tool is useful or harmful: surfacing a term the candidate earned, versus suggesting one they did not.
Read moreAI Engineering5 min
Personalisation and scale, in tension
Scale and personalisation pull against each other by definition. The useful question is which parts of a message can scale and which cannot.
Read moreAI Engineering5 min
Automating the follow-up, safely
Most of this system is not about sending. It is about knowing, accurately, what has already happened.
Read moreAI Engineering5 min
Automating a message worth reading
Recruiters decide in about four seconds. Almost everything a generated email wants to say is spent after that decision is made.
Read moreAI Engineering5 min
A CRM for a job search
Every job search tracker dies the same way: manual entry. If the user has to maintain it, they will not.
Read moreAI Engineering5 min
Finding a contact, legitimately
Guessing an address from a name pattern is the standard approach, and it is wrong on three separate counts.
Read morePut the advice to work
Reading about ATS screening helps. Running your actual resume through one helps more.
New guides added regularly