A critical CVE drops on a Friday afternoon. The old playbook kicks in: open a ticket, wait for a developer, and hope the patch lands by Monday. Or at least before an attacker does. AI breaks this playbook because now attackers move in hours. In fact, the current time to exploit sits at -7 days, which means attackers are finding and weaponizing vulnerabilities before the CVE is even disclosed.
When a vulnerability lands, someone has to fix it. How that fix happens has moved through three maturity stages: manual patching, automated patching, and now, agentic remediation. Each one changes who does the work and how much work the machine takes on.
Manual patching puts every step on a person. Automated patching is the practice of applying known fixes on a set schedule without a person running each update by hand. Agentic remediation goes further: AI agents find the exploitable flaw, generate and test a fix, and explain the change, with a human approving what ships. This article compares the three, with a focus on what changes when AI writes the fix.
The old model: manual patching on a compliance clock
Manual patching runs on a calendar, not on risk. A scanner produces a list, the list becomes tickets, and the tickets wait for a developer with time and context. The clock that drives the work is a compliance deadline: patch criticals within a set window, report the numbers, and move to the next cycle.
Even the most stringent compliance frameworks today, like FedRAMP, set this compliance clock at 30 days for critical or high vulnerabilities.
This model made sense when software shipped in quarters and attackers moved in weeks. It breaks under modern conditions. The backlog grows faster than any team can drain it, developers stall on code they did not write, and the compliance clock rewards closing tickets over reducing risk. A team can hit its patch SLA and still leave the one reachable, exploitable flaw open, because the clock counts tickets, not exposure.
What agentic remediation changes
Automated patching removes the manual step of applying a known update, reducing backlog. Patches automatically land on a schedule without a person pushing each one.
However, it shares manual patching’s blind spot. It applies fixes without checking which ones matter. It also does not check if they break the build. Automated patching is faster manual patching, not a different model.
Agentic remediation is the different model. Instead of applying a fix on a schedule, AI agents reason about the problem. They find which vulnerabilities are reachable, generate a candidate fix, test it through more agents, and explain the change before a human approves it.
The shift is from applying to reasoning. Manual patching asks a person to do the work. Automated patching asks a script to repeat a known action. Agentic remediation asks a system to decide, act, validate, and account for what it did.
Three changes define the shift:
- Reachability replaces the compliance clock as the thing that drives priority, so the system fixes what an attacker can hit first.
- Recursive validation replaces the single manual review, so one agent proposes, another tests, and a third confirms no new risk.
- An audit trail replaces the closed ticket, so every change carries its reasoning.
Visit our Agent Kit docs for a catalog of ready-to-use security agents that run inside your AI coding assistant.
Agentic remediation vs. manual patching
The three models line up on the same set of questions:
Manual patchingAutomated patchingAgentic remediationWhat drives priorityCompliance deadlineUpdate scheduleReachability and exploitabilityWho applies the fixA developer, by handA script, on a scheduleAn agent, for human approvalJudgment about which fix mattersHuman, case by caseNoneReachability-gatedBreaking-change checkManual testingTest suite, if configuredUpgrade impact analysisValidationOne manual reviewWhatever the pipeline runsRecursive: propose, test, confirmRecordA closed ticketA merge logA signed, auditable trailFailure modeBacklog growsShips fixes that break or missBlocked before it shipsSpeedBacklog timeSchedule timeLoop time
Read down the agentic column and the pattern is clear: the machine takes on the repetitive work, and the human keeps the decisions that need judgment.
Where humans stay in the loop
Agentic remediation does not remove the human. It moves the human up the stack, from doing the work to governing it. The system drafts and validates the fix. The person sets the guardrails and approves what ships, and that split is the point.
Humans stay in the loop in three places. They set policy: which systems an agent can touch, and which changes need sign-off before they merge. They approve high-risk changes: a fix to production code, or a change on a business-critical path, still passes through a person. And they own the exceptions: when the system flags a case it cannot resolve, a human decides.
Upgrade impact analysis assesses feeds that decision by showing the blast radius before anyone approves the merge.The goal is not full autonomy. A human who reviews validated, explained fixes covers far more ground than a human who writes each patch by hand, and the audit trail ties every automated action back to a reason. That is what changes when AI fixes vulnerabilities: not the absence of people, but the shift of people from labor to judgment.
FAQ
What is automated patching? Automated patching applies known software fixes on a set schedule without a person running each update by hand. It removes the manual apply step, which speeds up patching, but it does not judge which fixes matter or whether a fix breaks the build.
What is the difference between automated patching and agentic remediation? Automated patching repeats a known action: apply the update on a schedule.
Agentic remediation reasons about the problem: it finds what is reachable, generates a fix, validates it through several agents, and explains the change. One automates the hands. The other automates the judgment, with a human approving the result.
Does agentic remediation replace developers? No. It moves developers from applying fixes to governing them. The system drafts and validates the fix, and the human sets policy, approves high-risk changes, and owns the exceptions.
Is agentic remediation safe for production? Yes, with controls in place. Humans approve changes to production and business-critical paths, recursive validation tests each fix before it merges, and an audit trail records the reasoning behind every change, so a person can review any action.
What's next?
When you're ready to take the next step in securing your software supply chain, here are 3 ways Endor Labs can help:
