Most organisations have spent years attempting to eradicate friction from work. Fewer clicks. Sooner approvals. Automated handoffs. Much less ready.
AI-driven workflow automation is the following step in that story. It doesn’t simply automate duties; it automates choices and actions that used to require an individual within the loop. That effectivity is actual. It is usually the place a refined safety regression begins.
Human friction was by no means solely an inconvenience. It was a management floor.
Friction did greater than gradual folks down
In mature techniques, “friction” reveals up as small pauses and checkpoints that drive intent to be specific. The pause earlier than a payout is processed. The second a assist agent reads the ticket historical past, they will problem a refund. The second set of eyes on an entry grant. The acutely aware determination to run a manufacturing change.
These moments weren’t good fraud prevention. However they created three safety properties which are straightforward to underestimate.
They’ve restricted velocity. A human can solely achieve this a lot in an hour, and that naturally capped harm.
They offered a sanity test. Folks discover when one thing feels off, even with out formal guidelines.
They served as an authorisation checkpoint.
When AI removes friction, it removes these properties until you exchange them intentionally.
The place friction used to reside in workflows
Safety groups typically consider controls as coverage and know-how. In observe, most of the strongest controls have been embedded in how work occurred.
Assist groups had escalation norms. Finance groups had twin management for delicate actions. Engineering groups had code evaluate, change administration, and deployment gates. Operations groups had runbooks that required somebody to learn context earlier than performing.
Even casual habits created guardrails. Folks requested a colleague earlier than doing one thing uncommon. They hesitated when the request felt unusual. They slowed down when the results have been unclear.
AI brokers and automation pipelines don’t hesitate. They execute.
Additionally Learn: Shadow automation: The brand new insider threat
What AI modifications
AI doesn’t merely do the identical work sooner. It modifications the distribution of error and abuse.
First, it compresses time. Actions that when took minutes of human effort now take seconds. That reduces the window for detection and intervention.
Second, it will increase throughput. An attacker who compromises an account was once constrained by what they might do manually with out elevating suspicion. With automation, the identical compromise can set off large-scale actions rapidly: refund loops, payout reroutes, mass information exports, privilege grants, or ticket closures that cover fraud.
Third, it blurs accountability. When an motion is taken “by the assistant,” who authorised it? The consumer? The mannequin? The workflow designer? The supplier? Ambiguous possession weakens response and root trigger evaluation.
Fourth, it turns pure language into an execution interface. In case your platform permits “ask the agent to do it,” you might have successfully created a command layer that have to be secured like an admin console, even when it appears like chat.
The brand new failure modes that seem like regular operations
A key problem with AI-driven automation is that abuse can look reputable. The actions are carried out utilizing legitimate credentials, by means of authorised instruments, inside regular techniques. They don’t resemble basic intrusion patterns.
The failure modes are sometimes business-logic failures, not vulnerability failures.
An agent drafts and sends emails or invoices to the unsuitable events as a result of the immediate context was polluted.
An automation chain escalates privileges as a result of it was designed to “unblock” customers and by no means acquired a decent coverage boundary.
A mannequin makes a believable however unsuitable determination, and since the system is optimised for velocity, that unsuitable determination turns into repeated behaviour.
That is what the lack of friction actually means. The system turns into environment friendly at scaling each good outcomes and dangerous ones.
Friction as a product function, not a compliance tax
The widespread response is to bolt on extra approvals. That normally fails as a result of it treats friction as a penalty somewhat than a design primitive. Customers bypass it, groups disable it, and the organisation returns to shadow workflows.
A greater strategy is to deal with friction as an adaptive, intentional a part of the product.
The query isn’t “how will we deliver again guide steps?” It’s “the place do we’d like deliberate slowdown and specific intent as a result of the draw back is excessive.”
Additionally Learn: Automation: Are you main or lagging within the race?
Designing for intentional friction
Begin with a easy segmentation: low-risk actions, medium-risk actions, and high-risk actions.
Low-risk actions ought to keep quick. AI can and will take away friction the place the impression is reversible and bounded.
Excessive-risk actions want a special posture. If an motion strikes cash, modifications entry, exports information, modifies insurance policies, or impacts many customers, it ought to require a number of of the next: stronger authentication, a second approval, a delay window, or a tough coverage test.
The purpose is to not make the consumer endure. The purpose is to drive readability in the mean time of consequence.
There are sensible patterns that work effectively right here.
Step-up verification for delicate actions: If the agent is about to reroute payouts or grant admin entry, require the next assurance step, even when the consumer is already logged in.
Two-person integrity for irreversible actions: When the blast radius is excessive, require a second human approval. That is particularly essential for modifications that have an effect on many accounts or a monetary circulation.
Deliberate delay for sure actions:. A brief maintain on first-time withdrawals, payout vacation spot modifications, or mass exports creates time for detection and cancellation.
Coverage engines that gate execution: Don’t let the agent “determine” coverage. Let it suggest, then implement by means of deterministic guidelines.
Reversibility by design: Make it straightforward to undo what the agent did. If rollback is tough, friction have to be greater.
Restoring the security alerts people used to offer
Human friction additionally offered context. Folks requested, “Does this make sense?” AI wants equal guardrails.
Require the agent to indicate its foundation for motion in a structured method: what inputs it used, what it intends to do, and what techniques it would contact. Not as a protracted rationalization, however as a transparent motion abstract {that a} human can approve rapidly.
Log intent, not simply outcomes. Conventional logs file that an API name occurred. AI workflows ought to file why it occurred, what immediate or rule triggered it, and what information was referenced. With out that, the incident response turns into guesswork.
Add circuit breakers. If the system sees uncommon velocity or repeated failures, pause automation and require human evaluate.
Additionally Learn: AI workflow instruments are powering automation. However are they safe sufficient?
The management determination behind all of this
The uncomfortable reality is that friction is usually eliminated as a result of organisations optimise for velocity with out pricing in the price of failure. AI makes that tradeoff sharper.
If leaders need AI to execute, not simply advocate, they’re accepting that the platform is shifting from “assistive” to “autonomous.” That calls for the next bar for controls, monitoring, and accountability.
You don’t have to worry automation. You do have to be sincere about what friction was doing for you, and you must reintroduce it the place it functioned as a safety function.
Closing
AI-driven workflow automation is highly effective as a result of it eliminates human pauses. These pauses weren’t simply inefficiency. They have been fraud detection, sanity checks, and authorisation moments woven into each day work.
The following section of safe product design isn’t about selecting between velocity and security. It’s about constructing techniques the place friction is intentional, proportional, and triggered by threat.
—
Editor’s word: e27 goals to foster thought management by publishing views from the neighborhood. You can even share your perspective by submitting an article, video, podcast, or infographic.
The views expressed on this article are these of the creator and don’t essentially replicate the official coverage or place of e27.
Be a part of us on Instagram, Fb, X, and LinkedIn to remain related.
The submit AI-driven workflow automation and the lack of human friction appeared first on e27.
