Blog - dtechsystems.co
AI Productivity Bottlenecks: Why Faster Work Does Not Always Mean Faster Delivery
AI productivity bottlenecks can leave a business waiting for the final result even when individual tasks become faster. A team may produce more code, draft more proposals, or prepare more content, yet deliver little additional work because review and acceptance cannot keep pace.
For business leaders, the practical question is where work goes after AI produces its first output. That is the point at which a productivity claim needs to become an operating discussion.
What the research examined
In their working paper, Artificial Intelligence in the Firm: Bottlenecks in Software Production, Fiona Chen and James Stratton study software engineering using data from an engineering analytics platform. The version reviewed here, dated 9 November 2026, covers 445 million work events involving 720,542 workers at 655 firms.
The authors use differences in the timing of firms' AI adoption to estimate effects. This is an observational research design whose interpretation depends on assumptions about the comparability of earlier- and later-adopting firms. It is not a randomised trial of ClickUp or a study of Dtech clients.
The paper distinguishes intermediate coding activity from final software output. That distinction matters: producing more material is different from getting reviewed, accepted work into use.
More activity did not produce proportional delivery gains
For coding assistants, the authors estimate a 30.4% increase in lines of code 12 months after adoption. Estimated effects on resolved issues and epics were much smaller. For coding agents, the paper estimates larger increases in coding activity and smaller increases in final output. The agents' effect on resolved issues is described as borderline statistically significant.
The review stage helps explain the gap. After agent adoption, the authors estimate that submission-to-approval time increased by 56.1%, alongside longer waiting and active review. These are estimates from the studied setting, not universal performance forecasts.
The paper also reports no detectable employment effect in its empirical analysis, with uncertainty around the estimates. Its longer-term employment and output scenarios are model simulations rather than observed outcomes.
A practical implication for ClickUp workflows
The study did not test ClickUp. Dtech's practical application of the bottleneck idea is to inspect the whole delivery path before expanding AI-assisted production.
Consider a hypothetical content team that can prepare ten drafts each week but can review only four. Increasing drafting capacity to twenty does not, by itself, increase publication capacity. It can instead create a larger queue of unfinished work.
Map the process through drafting, readiness for review, active review, revisions, and acceptance. Record the accountable reviewer and the information they need to decide. Keep the distinction between waiting and active work visible, even if the first version uses a simple manual measurement.
ClickUp can organise the associated tasks and owners. A useful configuration follows the team's actual approval process; it does not assume that moving a task to a particular status enforces an approval control automatically.
Measure the complete outcome
During an AI pilot, track four things: accepted output, elapsed delivery time, review effort, and rework. Compare similar work over a defined period and record changes in staffing or complexity that could affect the result.
If drafting gets faster while acceptance slows, investigate the queue. The reviewer may lack context, the output may need more correction, or too much work may be entering the stage at once. Each cause calls for a different intervention.
Try a bounded change, such as a clearer brief, smaller review batches, or protected reviewer time. Check the result before increasing upstream output again. Review quality remains part of the outcome; a shorter queue is not useful if acceptance becomes unreliable.
Choose the next improvement around the constraint
The research offers a reason to be precise about productivity. Businesses need useful work completed and accepted, alongside the ability to produce it efficiently. A faster task can support that goal when the surrounding process is ready.
Before adding another AI tool, identify where the work waits today, who can move it forward, and which change would help that stage. Then evaluate the result through the complete workflow.
Source: Chen and Stratton, working-paper version dated 9 November 2026, abstract and sections 1, 4–6, especially introduction pp. 3–6. The authors' public version is linked below and may be updated independently.
Sources and further reading
Build a workflow your team can operate
Dtech helps teams map processes, configure ClickUp and support adoption. Explore our ClickUp implementation services or book a call to discuss your workflow.
Finding Operational Waste: A Proven Framework
Finding operational waste is hard because most teams never look for it. They either don’t measure where the week goes, or measure it once and never act on it. The result is the same: hours leak out through coordination overhead that every person assumes is just “how work works.” It isn’t. Here’s how to see it, size it, and start getting it back.
01 The 40-hour illusion
Ask any team lead how their team’s week was spent and you’ll get a confident answer that is almost entirely wrong. Not because they’re careless, but because operational waste is distributed: fifteen minutes here, a forty-minute hunt there, a weekly report that takes half a day to assemble.
Research on knowledge work keeps landing on the same uncomfortable number: roughly 40% of the week goes to “work about work” — coordination, status reporting, searching, and switching — rather than the skilled work people were hired for. On a team of eight, that’s more than three full-time salaries spent on overhead.
Finding operational waste starts with naming the leaks. That’s what the framework is for.
02 The Five Leaks framework for finding operational waste
Every hour of operational waste we’ve audited fits one of five categories. Each has a diagnostic question. If your team answers it instantly with a specific example, that leak is real.
L1 Assembly overhead
Meetings that exist to coordinate rather than decide. They multiply because nobody owns the decision, so everyone gets invited “just in case.”
Diagnostic Did this week’s recurring meetings each produce a decision, or just schedule the next one?
L2 The status tax
Reporting on work instead of doing it. Decks, digests, and standups compiled by hand from tools that already contain the answer.
Diagnostic How long did your last status report take to build, and how many people actually read it?
L3 The hunt
Searching for the file, the brief, the latest version, the person who knows. Work scattered across tools guarantees this leak.
Diagnostic How many different places would you check to find the current version of your most important document?
L4 The rework loop
Doing it twice because context was lost between people, or the brief lived in someone’s head and came out wrong.
Diagnostic How much of this sprint is fixing or redoing last sprint’s work?
L5 Context switching
App-hopping, pings, and interruptions. Each switch costs more than the interruption itself: it costs the ramp back into focused work.
Diagnostic How many tools did each person on your team touch before lunch today?
03 The 5-minute audit
Before you measure anything precisely, run this. Tick every item where you can point to a specific example from the last two weeks. No estimates, no “probably”: a real instance or nothing.
0 of 6 ticked · be honest, last week counts
Three or more confident ticks means your operational waste is structural, not incidental. That’s normal. It’s also fixable, but only if you target it rather than “working harder,” which is how teams usually respond and why nothing changes.
04 Operational waste vs. real work
The goal of finding operational waste is never a calendar with zero meetings or a team that never talks. The framework targets coordination cost, not coordination itself. Cut the left column ruthlessly; protect the right one at all costs.
Operational waste
Real work (protect it)
Compiling a status report by hand every Friday
A recurring “sync” with no agenda and no decision owner
Searching across four tools for the latest version of a file
Reformatting the same update for three different audiences
Chasing an approval through a 40-message chat thread
Redoing work because requirements lived in someone’s head
Deciding, designing, writing, building, selling
Talking to customers and learning something new
Deep, focused work on the thing you were hired to do
Reviewing work to make it genuinely better
Debates that end in a real decision (even a long one)
Cutting a real discussion is not savings; it’s damage with a nicer chart. The test is simple: did it produce a decision, a deliverable, or a customer insight? If yes, it stays.
05 Measure before you fix
Don’t guess at the size of the leaks. Instrument them. Have the team track time for two weeks against the five categories, then attack the biggest one first. Two rules make finding operational waste actually pay off:
One leak at a time. Fixing everything at once is how audits die in a slide deck. The biggest leak alone is usually worth more than the other four combined.
Fix the system, not the people. If the status tax is your biggest leak, the answer isn’t asking people to report faster; it’s making the work visible so reporting disappears. Tools like ClickUp’s Time Tracking and live dashboards exist precisely so status compiles itself.
Try it: run your team through the Leak Ledger
Estimate honestly (per person, per week) and watch what the five leaks cost at team scale. The quarterly number is the one that gets budgets approved.
The Leak Ledger
TIME AUDIT · 1 WK
People on the team
−8+
Meetings that don't produce decisions6.0h
Compiling status updates by hand3.0h
Searching for files, context, versions2.5h
Rework from lost context or bad briefs3.0h
App-switching and interruptions4.0h
148hteam hours lost / week
46%of each 40-hour week
241team days lost / quarter
Structural problem. Your tools and rituals need redesign, not tweaks. Consolidate before you optimize.
FAQ: finding operational waste
What actually counts as operational waste?
Work about work: coordinating, reporting, searching, redoing, and switching. Anything that isn’t the work itself or a decision about it. A two-hour debate that ends in a real decision is not waste. A 30-minute meeting that schedules another meeting is.
How is finding operational waste different from time tracking?
Time tracking tells you where the hours went. This framework tells you which of those hours shouldn’t exist at all. You need the first to run the second, which is why measurement comes before fixing.
How often should we run the audit?
Quarterly. Waste creeps back in every time the team grows, a tool gets added, or a process changes. The checklist takes five minutes; the fix for the biggest leak usually takes a sprint.
What’s a realistic target?
Most teams can recover 15 to 25 percent of the week within one quarter by killing their single biggest leak first. Trying to fix all five at once is how audits die in a slide deck.
Do we need new tools to fix operational waste?
Usually the opposite: fewer tools, better configured. Consolidating tasks, docs, goals, and chat in one platform like ClickUp removes most of the hunt and a big chunk of the status tax, because the work and the reporting live in the same place.
AI Agents vs Automation: 5 Critical Differences Businesses Must Know
When it comes to AI agents vs automation, most businesses are asking the wrong question. They’re not asking “which one do we need” — they’re asking “how do we get one of whatever everyone else is calling AI.” That’s how companies end up overpaying for intelligence they don’t need, or underbuilding a process that actually required judgment all along.
Every product update this year ships with the same headline: “now powered by AI agents.” Every vendor pitch leads with autonomy and intelligence. If you’re trying to figure out what your business actually needs, the noise has gotten loud enough to bury the one question that matters. This guide breaks the AI agents vs automation comparison down properly — the definitions, the cost implications, the common mistakes, and the practical framework you can apply to any process in your business starting today.
1. AI Agents vs Automation: The Core Definition
The starting point of any AI agents vs automation decision is understanding that these are not two versions of the same thing. One isn’t a “better” upgrade of the other. They are built to solve fundamentally different kinds of problems.
Automation is deterministic. It runs on a fixed set of rules: if X happens, do Y. There’s no interpretation, no judgment, no “it depends.” You define the trigger, you define the action, and it fires the same way every single time, regardless of context.
In the AI agents vs automation comparison, the distinction comes down to one line: automation executes, agents decide. Once a business internalizes that, the conversation stops being “which one is better” and becomes “which parts of our workflow actually need judgment, and which parts just need to run reliably.”
2. AI Agents vs Automation: Cost and Complexity
Cost is where the AI agents vs automation decision has real financial weight, and it’s where a lot of businesses get burned.
Automation is cheap to build and cheap to run. It’s logic, not reasoning — a rule engine executing the same instruction every time it’s triggered. There’s no ongoing “thinking” cost, no token spend, no variability in output. Once it’s configured correctly, it just runs.
AI agents cost more, because you’re paying for judgment, not just execution. Every decision an agent makes involves interpreting context, which means compute, and which means a price tag that scales with how much reasoning you’re asking for. That’s not a reason to avoid agents — it’s a reason to be deliberate about where you deploy them.
This is where businesses tend to go wrong in one of two directions:
Overbuilding: Deploying an AI agent on a task that a simple rule could have handled. If the correct action never changes regardless of context, you’re paying agent-level cost for automation-level work. That’s money and complexity spent on nothing.
Underbuilding: Forcing a static, rule-based workflow to handle a task that genuinely requires judgment. The rule can’t account for the exceptions, the edge cases, the “it depends” scenarios — so it either breaks, or it quietly produces the wrong outcome without anyone noticing until it’s a bigger problem.
Common examples of automation:
- Move a task to “Done” automatically once every subtask closes
- Route a new support ticket by keyword or category
- Send a Slack alert when a deadline is 24 hours out
- Assign a new lead record to the right rep based on territory
3. AI Agents vs Automation: Where Companies Get It Wrong
The most common mistake in the AI agents vs automation debate isn’t picking the wrong tool — it’s skipping the workflow audit entirely.
Here’s the pattern: a business sees a competitor announce an “AI agent.” They don’t want to be left behind. They buy or build an agent and point it at a process that was never clearly defined in the first place — inconsistent task ownership, messy or incomplete data, no shared definition of what “done” actually looks like across the team.
The agent doesn’t fix any of that. It can’t. An agent making decisions inside a broken workflow just makes bad decisions faster and with more apparent confidence. Nothing has actually been solved — the chaos has just been automated and given a more convincing voice.
Only after answering those questions do they know where automation belongs and where an agent belongs — and in what order to build them.
4. AI Agents vs Automation: Sequencing Matters
There’s a sequencing mistake baked into most AI adoption efforts: companies try to deploy agents before they’ve automated the basics. That’s backwards, and it’s an expensive way to learn the lesson.
If your task statuses aren’t updating reliably, if your team is manually reassigning work that should route itself, if reminders are falling through the cracks — none of that gets fixed by adding an agent on top. It gets fixed by automation. Clean, rule-based automation is the foundation. It’s what makes your underlying data trustworthy and your workflow legible enough for an agent to operate inside without immediately going off the rails.
Get the automation layer right first. Then decide, deliberately, where judgment genuinely needs to enter the picture — and that’s where an AI agent belongs in the stack.
5. AI Agents vs Automation: What This Looks Like in ClickUp
Inside, Clickup the AI agents vs automation distinction isn’t theoretical — it’s built directly into how the platform is structured.
Automations handle the deterministic layer: status changes, task assignments, recurring work, due-date reminders, and notifications when something needs attention. None of it requires interpretation. None of it should cost you agent-level compute or complexity. It’s rules, running reliably, in the background, exactly the way they were configured to.
Super Agents handle the judgment layer. A new lead comes in — a Super Agent researches it, enriches the record with relevant context, and flags whether it’s genuinely worth a rep’s time before anyone manually touches it. A time-off request comes in — a Super Agent checks it against team capacity and existing coverage instead of someone manually cross-referencing five different calendars. Across a busy workspace, instead of a flat, undifferentiated list of a hundred open tasks, an agent can surface what actually needs attention right now and explain why.
You don’t have to pick one paradigm for your entire business. The AI agents vs automation line gets drawn task by task, process by process, inside the same workspace.
The One Question to Ask Before You Buy Anything
Before adding any AI tool — agent or otherwise — to your stack, there’s one question worth asking about the specific task in front of you:
Is this a decision, or is this a rule?
If it’s a rule — if the right action is always the same action, regardless of context — automate it. Do it now. It’s cheap, it’s fast to configure, and it removes the manual work without adding unnecessary cost or complexity.
If it’s a decision — if the right action genuinely depends on context that changes case by case — that’s where an AI agent earns its place. That’s the work worth paying for reasoning to handle.
Everything else — the tools, the hype, the “we’re an AI-first company” branding — is noise dressed up as strategy. The businesses that actually benefit from this moment aren’t the ones with the most AI. They’re the ones who took the time to understand the AI agents vs automation distinction and built accordingly, task by task.
FAQs
Is an AI agent just advanced automation?
No. Automation runs fixed rules with no interpretation — the output never changes for a given trigger. AI agents make decisions based on context that changes case by case. In the AI agents vs automation comparison, they’re built to solve different kinds of problems, and treating one as a simple upgrade of the other is where most AI adoption mistakes start.
Should a business start with automation or AI agents?
Automation first, always. It builds the clean, reliable workflow foundation an AI agent needs in order to make good decisions on top of. Deploying agents before automating the basics means asking a reasoning system to work with incomplete or inconsistent data.
Can automation and AI agents work together?
Yes — and in most real workflows, they should. ClickUp Automations handle the rule-based layer while Super Agents handle the judgment layer, inside the same workspace, so teams don’t have to choose one paradigm across the whole business.
How do I know which tasks need an agent instead of automation?
Ask whether the correct action ever changes based on context. If it’s always the same response to the same trigger, it’s a rule — automate it. If the right response depends on details that shift case by case, it’s a decision — that’s where an AI agent belongs.