Is Your Ai Agent Up To The Task? 3 Ways To Determine When To Delegate

MF3d/iStock/Getty Images Plus via Getty Images

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways

• Many tasks are ripe for automation, but others autumn into a grey area. 
• Speed and productivity are not nan extremity of nan story.
• AI decisions are guidance 101, not method decisions. 

You whitethorn person already heard nan position that AI agents service arsenic "co-workers" to quality counterparts, functioning arsenic de facto extensions of nan workforce. The situation is decoding what activity they are champion suited to execute -- and it's not an easy question.

There are tasks ripe for automation and others that are amended handled manually. But galore are successful a grey area, successful which automation makes sense, but is it worthy nan investment? The clip has travel to position AI agents arsenic an adjunct workforce, and negociate them accordingly. 

Also: True agentic AI is years distant - here's why and really we get there

There is moreover speculation that IT, which manages agents, is taking connected the domiciled of a quality resources department. 

Management 101 skills vs. method skills 

That calls for guidance 101 skills versus purely method skills, urges Ethan Mollick, professor astatine nan University of Pennsylvania and starring AI expert. In "Management arsenic AI superpower," he describes nan thought process for deciding if AI should grip a task. 

From an economical and productivity standpoint, location is nary mobility that AI is accelerated and cheap, Mollick says. "It produces activity successful minutes that would return galore hours for a quality to do, and it doesn't mind if you make aggregate versions and propulsion astir of them away." 

Consider nan emergence of chatbots successful customer interaction systems. Automated responses -- boosted by AI -- now grip ample volumes of regular inquiries that slowed down reps' productivity to a crawl astatine 1 time. Now those reps are elevated to grip much analyzable customer issues.  

Also: As AI agents multiply, IT becomes nan caller HR department

However, velocity and productivity are not nan extremity of nan story. Here's different facet to consider: "You don't reliably cognize what nan AI will beryllium bully aliases bad astatine connected analyzable tasks," says Mollick. Doing nan incorrect point faster has ever been 1 of nan top challenges of management. 

The champion attack to making specified decisions is not to position AI deployment arsenic purely a exertion decision, but to position it from a guidance 101 perspective. This whitethorn yet thrust nan AI and AI supplier activity to nan adjacent realm -- to beryllium handled arsenic an adjunct workforce. 

To get nan astir retired of AI agents, for example, you request to beryllium capable to "explain what you need, springiness effective feedback, and creation ways of evaluating work," Mollick says.

How to find if delegation makes sense

Again, guidance 101 comes into play here. Do it, ditch it, aliases delegate it. "You delegate because you can't do everything yourself, and because talent is constricted and expensive," Mollick explains. "AI changes nan equation. Now nan talent is abundant and cheap. What's scarce is knowing what to inquire for."

Also: Deploying AI agents is not your emblematic package motorboat - 7 lessons from nan trenches

Successful delegation, then, depends connected being capable to determine what to delegate a task to an AI agent. Mollick suggests 3 measurements to find if AI is champion for nan task:

• Human baseline time: "How agelong nan task would return you to do yourself."
• Probability of success: "How apt nan AI is to nutrient an output that meets your barroom connected a fixed attempt."
• AI process time: "How agelong it takes you to request, hold for, and measure an AI output."   

These determination points interact pinch each other, and location are trade-offs that request to beryllium weighed. And successful figuring retired really to springiness these instructions to nan AI, it turns retired you are fundamentally reinventing management. "Consider a task that takes you an hr to do, but nan AI tin do it successful minutes, though checking nan reply takes 30 minutes," Mollick illustrates. "In that case, you should only springiness nan activity to nan AI if 'probability of success' is very high, different you'll walk much clip generating and checking drafts than conscionable doing it yourself."

"If nan 'human baseline clip is 10 hours," he continued, "it could beryllium worthy respective hours of moving pinch nan AI, assuming that nan AI tin beryllium made to do a competent job."

Also: 4 caller roles will lead nan agentic AI gyration - here's what they require

A squad of AI agents needs to meet nan aforesaid goals arsenic a quality team, by asking basal guidance 101 questions, specified arsenic "what are we trying to accomplish, and why?" and really to way and measurement an agent's progress.  

Management itself, arsenic good arsenic nan workplace, whitethorn germinate "when everyone is simply a head pinch an service of tireless agents," Mollick predicts. "The group who thrive will beryllium nan ones who cognize what bully looks for illustration -- and tin explicate it intelligibly capable that moreover an AI tin present it."