Perla › Manifest › a job in AI
Manifesting a Job in AI the eval, then the write-up
You picture a headline-making launch. The actual day is smaller: an experiment that didn't quite work, a careful look at why, and a write-up that helps the next person not repeat the same dead end.
- A full script for a job in AI, written in the present tense and ready to read tonight.
- Shorter versions — one for falling asleep, one for the morning, and one line to carry with you.
- How to use it: No script builds the technical skill, the portfolio or the track record that gets an AI role offered — that is real study, projects and applications, often self-directed, and this page names that plainly rather than pretending a script substitutes for it.
- The common mistake: Picturing the prestige of the field, the headline launch, the label on a business card, while skipping the actual discipline, running the evaluation properly, reading the failure cases, writing them up honestly, sets you up to struggle in a real AI role even if the title sounds impressive.
The script
Read it slowly, in the present tense, as though it has already happened. It is written to be spoken aloud, but silently is fine.
I run the evaluation I set up yesterday and the results come back mixed, better on one measure, worse on another, and instead of being deflated by that I get curious about it, because mixed results are usually the interesting ones. I spend the morning going through specific examples where the model or the system got it wrong, actually reading them rather than just trusting the aggregate number, and a pattern starts to show itself that the summary statistic was hiding. In a meeting, someone asks a genuinely hard question about whether the approach even makes sense at a larger scale, and I don't pretend to know for certain; I say what the evidence so far actually supports and what it doesn't. I write up the findings plainly, including the parts that didn't work, because a write-up that only reports successes is close to useless to the next person. A colleague building on a different part of the system reads it and adjusts their approach because of something I found. Nothing about today made a headline. It moved a real, specific piece of understanding forward, carefully, and logged it clearly enough that it stays moved forward after I've gone home.
Shorter versions
Two for the ends of the day, and one line to carry through the middle of it.
The evaluation results are logged, mixed as they were, honestly written up. The pattern I found today is on record now, not just in my head. Nothing about tonight needs re-running. I let the day's uncertainty stay uncertain rather than forcing a tidy conclusion it doesn't deserve yet. That was real, careful work. It's allowed to rest unfinished.
I check yesterday's evaluation results with actual curiosity, not dread, even knowing they were mixed. Whatever hard question comes up today, I'll answer with what the evidence actually supports, not more than that. I am not chasing a headline. I am doing patient, specific work that moves understanding forward one honest write-up at a time.
Mixed results are the interesting ones — that took a while to actually believe.
How to use it
No script builds the technical skill, the portfolio or the track record that gets an AI role offered — that is real study, projects and applications, often self-directed, and this page names that plainly rather than pretending a script substitutes for it. Read this before a technical interview or while building a project, to settle into the ordinary discipline of the work, evaluating carefully, writing up honestly, rather than the hype around the field, and let the actual learning carry its own real weight.
The mistake to avoid
Picturing the prestige of the field, the headline launch, the label on a business card, while skipping the actual discipline, running the evaluation properly, reading the failure cases, writing them up honestly, sets you up to struggle in a real AI role even if the title sounds impressive. The other failure is rehearsing an interviewer's approval instead of your own comfort with genuine uncertainty. Keep the scene in the evaluation and the write-up, and let the actual study carry the rest.
Questions
Do I need a PhD to get a job in AI?
Not for most applied roles — a PhD matters more for certain research positions, while engineering and applied roles often value demonstrable project skill more.
Is the field too saturated now to break into?
It's genuinely competitive, but demand is broad and growing across many types of roles, not concentrated in one narrow entry point.
What actually helps most, courses or personal projects?
Most hiring managers in this field say a specific, well-documented personal project demonstrates more than a course certificate alone.
Have this written about you
The script above is written for anyone. Perla asks about your actual situation, writes it as a present-tense narrative and reads it back to you — in a calm voice or your own, recorded once. One ritual instead of ten apps.