You drop a three-screen Slack novel every time a cookie lacks HttpOnly. The bot writes that same Secure, SameSite, token-theft sermon before you type have you considered. The intern posts it first now. The thread reacts with the same damn emoji they used to give you.
Your busywork’s shelf life~14 months
Estimated AI takeover of 3 exposed tasks. Whole-job replacement: uncertain.
Opens a text draft and downloads your roast image. Add it with LinkedIn’s photo button, then review and post. If the text is blank, paste the copied roast.
An illustrative midpoint of 3 medium/high-exposure task windows: now–6 months → 3, 6–18 → 12, 18–36 → 27. High exposure gets weight 2; medium gets 1. Low-exposure tasks are excluded. This gives ~14 months from the assessment date, not a countdown to losing your job.
Calculation: (3 × 1 + 12 × 2 + 27 × 1) ÷ 4, rounded to the nearest month. The task assumptions below must hold.
The whole role is unlikely to vanish while live incident ownership, privileged access, and blame still attach to a named engineer. Models can draft triage and reports; they do not absorb the decision to shut something down.
Confidence: low. Timing is conditional; inferred duties may not match your actual workload.
Triage SOC alerts and noise
Now–6 months
medium exposure · Inferred from role
How AI takes over
AI models cluster log and alert streams, score likely false positives, and output a short queue; humans still confirm odd cases and context.
What has to happen first
Needs log connectors and a feedback loop on past true/false alerts before teams trust auto-close.
Draft incident writeups
6–18 months
high exposure · Inferred from role
How AI takes over
A copilot turns timeline notes and ticket text into a polished after-action; humans shrink to edits and sign-off.
What has to happen first
Adoption waits on template fit, confidentiality of tickets, and managers accepting machine-drafted narrative.
Hunt and review findings
18–36 months
medium exposure · Inferred from role
How AI takes over
Agents propose hypotheses from telemetry and code diffs and draft findings; humans still judge exploitability and business impact.
What has to happen first
Friction is environment-specific data, false-positive cost, and reluctance to let automation set severity.
What still needs you
Named ownership when a real incident needs a shutdown call and someone to take the blame.
Your next move
Document your shop’s false-positive patterns and run the copilot against last month’s tickets, then own only the calls it cannot make.
The score averages three task ratings: low = 20, medium = 50, high = 80. It's an illustrative index, not a probability of losing your job.
Preview the downloadDownload the image and attach it to your own post.
