top of page

18 Experts on What “Efficiency” Really Means in the Age of AI

Nov 13
13 min read

AI can produce more work in less time, but efficiency is about more than generating faster. Experts in the field explain how to focus saved time on original ideas, sound decisions, and measurable results. These insights show how teams can use AI without sacrificing quality, judgment, or impact.


  • Establish Release Criteria Early

  • Curate Machine Suggestions With Technical Discernment

  • Move Beyond Predictable Copy

  • Measure Total Effort to Acceptance

  • Evaluate Full Tasks Rather Than Draft Counts

  • Build Systems Through Tight Constraints

  • Prevent Issues at Their Source

  • Reach Trusted Conclusions With Fewer Alternatives

  • Spend Attention on Selected Material

  • Practice Restraint When Abundance Adds No Value

  • Prioritize Demand Ahead of Generation

  • Operationalize People-Machine Teamwork

  • Treat Verified Truth as the Metric

  • Tie Defensible Choices to Revenue

  • Redirect Saved Time Toward Original Impact

  • Reserve Personal Expertise for Final Pieces

  • Coordinate Concurrent Projects Without Quality Loss

  • Price Corrections Against Speed


Establish Release Criteria Early

Efficiency should be measured by the time it takes to produce a usable result, including review and decisions. I use AI to help prepare webinar presentations and speaker scripts. It saves considerable preparation time, while I direct the audience, teaching goals, examples and final wording. 


A useful comparison is the total time spent briefing, drafting, checking and correcting a deliverable. Generating ten versions is only helpful if it makes the next decision easier. Before using AI, define what finished means, name the reviewer and decide which errors must be caught before release. These comments were prepared with AI assistance using my firsthand webinar experience.



Curate Machine Suggestions With Technical Discernment

Efficiency used to mean typing faster. More code per hour, more configs per sprint. AI killed that definition and I don't think most people noticed yet.


I build data platforms for 1,500+ client accounts. I use AI tools every day. Here's what changed. AI generates 5 valid Kafka consumer configs in 10 seconds. All five compile, all five look reasonable. But picking the right one for your throughput and latency constraints, that takes 20 minutes of reading docs, checking partition counts, thinking about backpressure. Generation became free. Evaluation didn't.


Before this role I built a real-time analytics pipeline at a mobile gaming company (Cooking Diary, 22M+ downloads). Back then efficiency meant getting the Flink job written at all. Now AI writes it in seconds, the hard part is knowing if the watermark strategy holds under actual load.


Efficiency now means judgment. Knowing which AI output to keep, which to throw away. People who can't evaluate what AI gives them will drown in plausible garbage. The bottleneck moved from creating to curating, that's the whole story.


Timur Rakhmatullin, Senior Software Engineer, Softline Solutions


Move Beyond Predictable Copy

While I'm drafting a case study, AI can suggest 10 accurate headlines in seconds. Sometimes none of them lands.


That's by design. AI predicts the likeliest next word, so it hands you the most expected phrasing, and your competitors get something very close. Once competent copy becomes the default, it stops getting noticed.


The common reflex is to ask for another batch. Efficiency breaks right there. When one more round costs nothing, the decision keeps getting postponed.


I'd redefine efficiency as how fast you can move past the obvious to a line only you could write.


Julia Kuklenko, Content Manager, WiserBrand


Measure Total Effort to Acceptance

Efficiency should be measured by how quickly you reach a useful, verified outcome, including the work needed to check and apply AI's output. Generating ten drafts in a minute has little value if choosing, correcting and approving one takes longer than writing it yourself.


For an AI-assisted workflow, I would track three things: time from request to an accepted result, the proportion of results that need rework, and the human effort required per completed task. For writing, that means an accurate message approved for its intended audience. For analysis, it means a decision supported by checked evidence.


Set the acceptance criteria before generating anything. Then ask for the fewest outputs needed to make the decision, check them against those criteria, and stop when the task is complete. AI makes it easy to produce more material; efficiency comes from reducing the total effort between the original need and a result someone can actually use.


Rami Zwebti, AI Business Consultant


Evaluate Full Tasks Rather Than Draft Counts

Must be able to assess the amount of work needed to accomplish a task which is meaningful. With AI, you can always generate more content, but having the time to read it, filter out the content that's most important and act on it requires the same effort.


For example, I would include a full work flow: check, edit, and transmit. It does no good if a message to a patient takes just seconds to prepare if another staff member spends twenty minutes correcting or answering questions about it.


Define before resorting to AI. The output is what matters, so whom will you target, what are the limits, and what judgment should it help you make? Ask only as many questions as you can assess.


The real question is: will the work be completed earlier, to adequate quality and with less re-work? That keeps productivity associated with results, not with some number of bytes.


Sanju Zachariah, Software Specialist, Management Consult for IT Automation, IT Program Manager, Founder & President, Portiva


Build Systems Through Tight Constraints

Efficiency used to mean producing more output per hour. Now that output is effectively free, efficiency means how fast you can turn a good decision into a working system. Generating 20 options in seconds is not progress if a human then spends an afternoon reading them.


The better question is what the AI shipped, not what it wrote. I built batko.ai with Claude Code, and it grew to 12 apps, more than 4,000 pages and 442 API endpoints. The lesson that came with that speed is the other half of the new definition: I moved so fast I never stopped to look at how it deployed, and it turned out to be one monolithic build where a single push rebuilt everything. The mess compounds quietly when generation is instant.


So I would redefine efficiency as two things held together. First, decisions per hour, meaning you give the model tight constraints so it narrows your options rather than multiplying them. Second, the discipline to stop and fix the structure underneath before speed turns into drag. If your AI use is producing more for you to read rather than less for you to do, you have made yourself less efficient, not more.



Prevent Issues at Their Source

Efficiency should now be measured by the work that never had to happen, not by how much output you can generate. AI made producing text and options almost free, so volume is no longer a useful measure of anything. I saw this in my own inbox recently: 267 LinkedIn connection requests in one week, and about 90 percent of them were AI-written using the same formula. That is not efficiency. It is noise produced faster.


In ecommerce customer service the difference is easy to put a number on. Every support ticket takes a person 15 to 25 minutes to resolve and costs around $22 in human time. Using AI to write those replies faster saves a slice of that. Using it to spot the late shipment or failed payment early, and fix it before the customer writes in, removes the ticket and the whole $22 with it.


So my definition is this: an efficient team uses AI to remove the reasons work exists, then spends the time it gets back on decisions only people can make. If the main thing AI is doing for you is helping produce more drafts, more messages and more options, you have probably made the busy work faster rather than made the business better.


Jevon Le Roux, Co-founder & CEO, Keeyu


Reach Trusted Conclusions With Fewer Alternatives

Efficiency now means time to a decision you can trust, not how much you produce. When AI generates ten options in seconds, output stops being the scarce resource. Judgment is. A team that produces three times the drafts but spends twice as long checking them hasn't gained anything.


I see this in enterprise AI rollouts all the time. The projects that pay off measure the moment a person acts, not the moment the AI finishes. On one ServiceNow platform serving 85,000+ employees, we used AI to summarize long ticket histories. The win wasn't the summary itself. It was that engineers could start working on a case about 10 minutes sooner each time.


Same principle in security. Our AI remediation framework doesn't flood people with fixes. It ranks the options, attaches a confidence score, and sends one clear recommendation to the person who owns the system. Fewer options, faster decisions.


So the new questions are simple. How long does it take to verify this? How many options does someone really need? Who decides?


When generating an answer costs nothing, the only efficiency that counts is how fast you can trust one.


Mahesh Shivnatri, Technology Leader, Deloitte


Spend Attention on Selected Material

A friend's sourdough starter, which she feeds every morning before her coffee. She told me the bread takes about 10 minutes of real work spread over 2 days. Nobody calls that inefficient.


I write content at a firm that lines founders up with investors. AI hands me 15 versions of an intro paragraph in under a minute, which used to be my whole morning. Picking 1 of them still takes about as long as writing it did (maybe longer, honestly).


So I have stopped measuring efficiency by how fast something exists. It probably means fewer things finished and more of them read twice. If generating is free now, what exactly are you spending your own attention on?


Drushi Thakkar, Sr. Creative Strategist, Qubit Capital


Practice Restraint When Abundance Adds No Value

Efficiency used to mean producing more in an hour. That definition is dead. When AI can produce endlessly, the bottleneck moves to decisions, so efficiency now means deciding better in an hour. I've seen teams generate ten versions of everything and call it progress. It isn't. They've moved the cost from writing to reading, and reading ten AI drafts takes longer than writing one good one.


The underrated skill is restraint. AI earns its keep when options actually matter: exploring an approach you haven't tried, stress-testing a position, starting something you don't know how to begin. When you already know the answer, generating more is waste in a productivity costume. The human role doesn't shrink, it concentrates. The machine proposes, the person decides, and anything touching money or trust keeps a person in the loop.



Prioritize Demand Ahead of Generation

It's important to note that efficiency shouldn't be measured by what an AI can spout out in a short time but by how much useful work one has carried out. Getting 10 tasks done is of little value if only 2 were needed. More energy should be put into deciding what is worth doing before using AI to get these tasks done. Efficiency means using the time saved to work on what's impactful.


We work with eCommerce brands to determine what their customers are looking for to target the highest potential for traffic. For example. an eCommerce site may be able to generate hundreds of product descriptions using AI within minutes. But that's not necessarily efficient if nobody's really interested in those products. I much rather eCommerce sites figure out which product pages need improved, and use AI to help improve them.


Bram Vermolen, SEO Expert/Strategist, Space'M Online


Operationalize People-Machine Teamwork

Efficiency today should be defined by how well people collaborate with AI, not by how fast AI can produce text or options. Instant generation shifts value to human judgment, curation, and integrating AI output into real workflows. Teams should measure efficiency by clearer decisions, less rework, and smoother handoffs between people and AI. As I argue, the real advantage comes from operationalizing collaboration with AI rather than treating it as a novelty.



Treat Verified Truth as the Metric

Efficiency used to mean output per hour. It now means verified output per hour.


When generating text costs nothing, the bottleneck moves from producing to checking. I learned that on SmartKeys: a model drafts a review in minutes, and about a fifth of its statistics cannot be sourced. Every minute saved writing became a minute spent verifying, and verifying is slower, because you cannot skim it.


So I changed the measure. The valuable hour is one in which something true and specific left my hands: a number with a source and a date, an opinion I will defend, a decision with a log. Drafts and variations are no longer achievements. They are raw material, and raw material is free now.


The old efficiency rewarded people who could produce quickly. The new one rewards people who can tell, quickly, what is true.



Tie Defensible Choices to Revenue

In my team AI already does a lot of the heavy lifting. It clusters thousands of search queries in minutes, which used to take a specialist two days. It writes first drafts, helps us analyze dozens of competitor sites in one evening, and handles routine work like meta tags, alt text and translations. Because of that, we stopped measuring efficiency by how much gets produced. Anyone can get ten versions of a text in a minute now.


For us efficiency means how quickly a person gets to a decision they can defend, and whether that decision moves revenue. The final version of anything we publish is still written by a human, and the hours AI saves go into editing, fact-checking and reading the numbers.


That checking part matters more than people expect. For one of our roofing clients, leads that came from ChatGPT were showing up in reports as ordinary referrals until we rebuilt the attribution. If we had trusted the dashboard as it was, we would have made budget decisions based on the wrong picture. So my advice is to let AI take the volume and keep people responsible for judgment, facts and the brand, because taste and responsibility are the parts you can't hand over to a model.


Mariia Grachova, Co-Founder & CEO, Loona Agency


Redirect Saved Time Toward Original Impact

Efficiency should be measured by how much time you free to do higher-value work and improve outcomes, not just how fast you can generate options. After integrating AI tools in early 2023, we saw a 35 to 50 percent reduction in time spent on prospect responses, proposals, content ideation, drafting and other administrative tasks. 


We use AI to produce first drafts of messaging frameworks and content, which our team then refines to preserve authenticity and quality. Redefining efficiency means investing in team training and workflows that pair AI with human judgment so the time saved amplifies creativity and strategy, making impact and originality the true measures of success.



Reserve Personal Expertise for Final Pieces

Efficiency used to mean producing more. Now it means better decision-making.


AI cut my first drafts from about eight hours to under two. The easy mistake would have been to call that six hours saved. It wasn't. Most of that time was spent on the work AI can't do well: checking every claim, cutting what's generic, and adding first-hand experience that nobody else has.


That's where the value sits now. Words are cheap. A test, a number, or a mistake that only I can share is not.


So I measure efficiency differently. I don't ask how fast the draft came out. I ask how much of the finished piece only I could have written. If the answer is nothing, the piece isn't done, however quick it was.


To be clear, the time I save on drafting does let me publish more. I went from three or four posts a month to ten to fifteen. But the extra output only counts because my editing and shaping of the article didn't shrink. That's how I know I am still producing content that will still be useful to the reader.


My advice: let AI handle the blank page. Keep the judgment for yourself. Then share your work, see what lands, and keep refining. That's the part worth your hours.



Coordinate Concurrent Projects Without Quality Loss

I think people are confusing efficiency with how quickly they can produce something. Just because AI lets you create ten things in the time it used to take you to create one doesn't mean you're being more efficient. What happens if all ten get rejected by the client or don't produce the results they were supposed to? Now you have to spend even more time fixing everything.


Personally, AI has made me way more efficient, but I think that's because of how I use it.


Before AI, research was probably one of the most exhausting parts of my job. I'd spend hours on Google going through websites, reading articles that all seemed to say the same thing, and trying to find useful information. When I started using ChatGPT for research and writing social media campaigns, I was able to cut that time roughly in half.


But I think the biggest difference is how I manage my time, and that's actually something I learned working in food service.


When you're in a kitchen with a bunch of hungry customers, you can't put bread in the oven and just stand there waiting for it to finish. You put it in, figure out what else you can prepare while it's baking, and keep things moving. I've carried that mentality into pretty much everything I do.


That's how I use AI now. I might have ChatGPT generating something for one client while I'm in Figma designing for another. When it's finished, I go back, review what it produced, make corrections if needed, and let it work again while I return to the other project. Each project still gets the attention it deserves. I'm just not sitting around waiting for one thing to finish before starting another.


Where I think people get it wrong is expecting AI to do everything from start to finish. I've had situations where I could've made something in Figma in five minutes, but trying to get AI to produce the same result took twenty or thirty minutes of prompting and corrections. At that point, how is that more efficient?


I don't necessarily work fewer hours now. Sometimes I probably work more. But I'm not nearly as mentally exhausted because I can spend more of my energy on the actual creative work instead of all the preparation.


I guess that's how I'd redefine efficiency. It's not about producing as much as possible or doing everything faster. It's about making better use of your time without sacrificing the quality of what you're producing.


Azul Duran, Founder & Creative Director, Drift Studio


Price Corrections Against Speed

AI has made generating work almost free. The real measure of efficiency is how much of that work you don't have to undo.


We've spent decades measuring productivity by how quickly people produce things. AI changes that equation. It can generate a hundred ideas, write thousands of lines of code, or produce a polished strategy in seconds. But generating something and getting it right are two very different things. If the output is misleading or sends a team in the wrong direction, we've accelerated the creation of work rather than eliminated it. An AI hallucination becomes expensive when it sounds convincing enough that an entire team builds its work around it.


Efficiency has to include the cost of verification, correction, and rework. Speed matters, but it means very little if the result doesn't stand up to scrutiny. AI can lower the cost of producing an answer, but it doesn't lower the cost of believing the wrong one.


 
 

Personal. Private. Everywhere

Get the next generation of Lightkey and write smarter in every app.

For Windows · Free for 30 days · No sign-up or credit card

bottom of page