Skip to main content
Blog_New_Analyst_Learning_Curve.jpg
Company

The New Analyst Learning Curve: Training and Retaining Talent in the Autonomous SOC

By Arijeet Ghatak

Ask a senior analyst how they learned the job, and closing false positives rarely comes up. That work mostly built alert fatigue. The learning happened in rarer moments: the real hit, the escalation that became an incident, the Tier 2 analyst walking you through what you missed.

Agentic investigation now clears much of that queue. For coverage, that's a clear win. For your bench, it removes the place where people learned. Earlier this year, we mapped where the analyst role is heading. This blog covers the leader's side: how to get your people there without losing what the queue used to teach.

Security comes from habits repeated in everyday moments. Analyst judgment is built the same way, through reps. Naturally, if AI removes those reps, leaders have to schedule the practice back in.

The Autonomous SOC can relieve the grind that drives burnout, but only for leaders who deliberately maintain the learning that used to happen naturally. The grind was also the learning curve, and organizations must prove it that they haven’t lost it.

Two Talent Problems, One Shift

By taking over triage, AI now affects retention and the talent pipeline at once, and each problem feeds the other.

Start with retention: in addition to shift work, after-hours pages, understaffing, and pay, repetitive work has driven SOC attrition for years. AI is supposed to relieve that repetition, but the early data is mixed: in ISC2's 2026 Rethinking AI's Impact on Cybersecurity Roles report, 48% of professionals using AI say it lowered their stress, while 32% say it raised it. That second group is more likely to report spending extra time validating AI output and deciding when to trust it. Without structure, AI moves the stress into validation work.

The pipeline shows a similar split: the same report found 56% say AI has reduced the need for entry-level roles, while 53% say it is creating new entry-level opportunities. Both can be true: the old entry-level job is shrinking and a different one is taking its place. And AI skills now top the list of skills needs in ISC2's 2025 Cybersecurity Workforce Study, cited by 41% of respondents who reported a skills need. You can't hire your way out of a senior shortage if no one is coming up the ladder behind them.

The obvious objection: if AI handles triage, juniors don't need to learn it. ISC2's respondents disagree: 62% say AI has not reduced the need for foundational skills, versus 26% who say it has. Someone still has to recognize a wrong verdict, and that judgment used to come from triage.

The Capabilities At Risk

Reviewing an investigation is a different skill from conducting one. An analyst can become very good at evaluating an automated system's work without the independent expertise to see when that system is fundamentally wrong. The capabilities most at risk are the ones the old queue exercised without anyone planning it:

  • Approaching an unfamiliar problem with no verdict to start from
  • Forming competing hypotheses instead of testing the first one offered
  • Noticing missing or incomplete evidence
  • Recognizing when the premise of an investigation is wrong

Write these down. Every practice below should build at least one of them, and every metric should test whether they hold.

What the Job Is Now

Responsibilities are shifting at every level of the SOC.

  • Tier 1 analysts move from closing every alert to reviewing AI verdicts and escalating the ones that don't hold up.
  • Natural-language querying speeds up hunting, but writing and reading queries is how they learn the data: the schema, the parsing gaps, the missing logs.
  • The path out of triage runs toward detection engineering, where analysts write and tune the logic the AI depends on.
  • Workflow design and leadership reporting stay with SOC engineering, IR leads, and managers, who now also own the rules the AI works within.

Each new responsibility assumes judgment that analysts used to build by doing the old one. Training must supply it now.

Training: Rebuilding the Learning Curve

Training is also how autonomy gets earned, as we covered in our retrospective review of our Autonomous SOC Maturity Model. Partial Autonomy (Level 3) is built on how analysts conduct investigations independently, and how they review investigations: which AI results they accept, modify, or override. Trained analysts make their records trustworthy before the system acts alone.

Five practices a SOC manager can start this quarter:

  • Run blind investigations. Have analysts work a share of live alerts cold, before they see the AI's verdict. Replay past incidents the same way, with the verdict withheld. Then use the AI's evidence chain as the answer key. Additionally, expand peer reviews: Add in 5 to 10 AI-closed cases per analyst per week and run quality assurance. This keeps investigative skills sharp and teaches where AI is reliable and where it needs a human.
  • Schedule the reps. Purple teaming, atomic tests, replays of past incidents, and tabletop exercises put analysts in front of attacks where the answer is known. Put them on a fixed calendar so they survive a busy quarter. Emulated attacks also let you score what the AI missed and what the analysts missed, separately. Current threat intelligence keeps the techniques realistic.
  • Make the AI's work the curriculum. Have juniors trace agentic investigations back to the raw telemetry: what was gathered, what was compared, why the conclusion followed. In addition, review the misses: wrong verdicts expose exactly where the AI's reasoning broke. Ask why the AI missed and feed the answer back into your workflows.
  • Treat ownership as governance training. Stage people the way you stage automation: crawl, walk, run. Juniors start by reviewing verdicts, then own a low-risk automated workflow such as phishing response. From there, they expand by domain. Ensure they read generated queries, because they’re learning AI verifiability and governance.
  • Train analysts. Give each analyst a yearly rotation through incident response, threat hunting, or detection engineering. Teams without those functions in-house can co-investigate with an MDR or IR partner.

Retaining: Make the New Work Visible and Rewarded

Training without recognition just builds a better resume for your competitor.

  • Rewrite the career ladder. If promotion criteria still count alerts closed, you're rewarding work that AI now does. Promote on independent investigation performance, as well as hypotheses tested, AI misses caught, and detection coverage added.
  • Change what you measure. Measure independent detection rate on emulated attacks, and pair it with a rubric that measures blind investigation quality. In addition, run unassisted exercises on novel attack scenarios twice a year.
  • Protect the time you free up. If saved capacity is immediately backfilled with more queue work, you get the same burnout with better tools.
  • Give real governance authority. Analysts who set the boundaries the AI works within have a stake in the system, and that ownership is a reason to stay.

Training for the Work That's Here Now

In the Autonomous SOC Maturity Model, Partial Autonomy (Level 3) is where the work is now. Human oversight is moving from reviewing individual actions to governing a system's behavior within a defined scope. However, training programs can produce good-looking metrics while independent skill erodes for years.

As such, the idea of verifiability is more important than ever. Purple AI, for example, shows the evidence chain behind every verdict, so you can grade analysts against it, test it against emulated attacks, and see where each falls short.

Two questions belong on every security leader’s agenda:

  • Which capabilities are we preserving?
  • How do we test that analysts can perform independently? 

We believe that expansion from Partial Autonomy (Level 3) to High Autonomy (Level 4) should occur only when independent human capability is shown to hold. Governance precedes autonomy, and proven human skill is part of that governance.

Related Articles

Decorative background gradient

Subscribe

Get the Latest From the SentinelOne Blog