Reducing Cognitive Load in Intelligence Work: Tools, Tactics, and Tradeoffs

The Workers Gaining the Most From AI Report the Most Burnout

Among employees in the top quartile for self-reported AI productivity gains, 88% reported burnout, and they were twice as likely as their lower-output peers to be considering quitting, according to a July 2025 Upwork Research Institute survey of 2,500 workers. A BCG study published in Harvard Business Review in March 2026, covering 1,488 US full-time workers, found that 14% of AI users reported "AI brain fry," acute mental fatigue from using or overseeing AI beyond their cognitive capacity. Those who reported it were more likely to plan to leave their roles than those who did not (34% versus 25%).

BCG's authors separate the two conditions: brain fry is acute mental fatigue from overseeing AI, and burnout is chronic workplace stress. Neither study isolates intelligence or security professionals, so this piece tests both findings against what is documented for those teams and looks at what drives AI analyst burnout among them. For the general argument about why AI often leaves workers busier, see AI Was Supposed to Save Time. Why Are Teams Busier Than Ever?

Where Cognitive Load Comes From in Intelligence Analysis

Cognitive load in intelligence analysis builds from three sources: the volume of routine work, the number of tools an analyst moves between, and the work of producing the report itself.

Routine work and burnout. The Tines Voice of the SOC Analyst survey (2022) found that 71% of analysts reported some level of burnout, and that reporting, monitoring, and detection were the tasks that took up the most time. Most respondents were Level 2 and 3 analysts, and Tines reports they spend most of their time on front-line monitoring. In the survey's account, understaffing, overwork, and tedious tasks sit behind SOC analyst burnout.

Tool sprawl and context-switching. The same survey found that 53% of analysts used between 11 and 30 security products. Ophir, Nass, and Wagner's 2009 PNAS study found that heavy media multitaskers were worse at filtering out irrelevant information and organizing material in memory, even as they felt productive. That study examined media multitasking in a lab, so it is best read as an analogy for security tooling: switching between sources and consoles may carry a similar cost that people underestimate.

Report production. Intelligence work adds a second layer on top of triage: reports have to be drafted, sources validated, citations formatted, and findings reshaped for different audiences, often from the same research.

The Verification Burden: What Reviewing AI Output Costs an Analyst

This piece uses verification burden for the time and attention spent confirming that AI output is correct, a closely related mechanism that BCG's research calls oversight. An analyst reads a claim, judges whether it holds, and moves between the output and the source material to check it. That back-and-forth is cognitive work even when no task list records it.

AI-assisted reporting looks like less work on paper because drafting time drops. If the analyst can't see where each claim came from, though, they have to rebuild the evidence trail themselves: opening tabs, re-reading sources, and holding the output in working memory while they do. Say a draft states that a regional outlet reported a new sanctions designation. The analyst has to find that article, confirm the date, and check whether the draft dropped a qualifier, all while keeping the rest of the draft in mind. The hours saved on drafting can go straight into checking.

Tools that promise to cut alert volume or summarize threat intelligence can add this burden too: unless each claim traces back to a source, the analyst's new job is judging the system's output. For a sense of where adoption stands, ISC2's 2025 AI Pulse Survey found that 30% of cybersecurity teams report having successfully integrated AI into their operations.

BCG's research found that burnout was lower among workers whose AI use took over repetitive, tedious tasks, and that brain fry decreased when managers provided training and support on AI tools. How much load AI adds depends heavily on what it is used for and how its output is set up for review.

Four Ways to Reduce Cognitive Load in AI Workflows

Reducing cognitive load in AI workflows comes down to four design decisions, each aimed at a different source of friction. All four leave judgment with the analyst and protect time for it.

1. Fewer Tools, Fewer Seams
Addresses: tool sprawl and context-switching. Moving from a search platform to a browser to a translation tool to a document editor forces a context reset at each stop, and the working memory needed to carry context across tools that don't share it is part of the cost. Consolidating inputs into one workspace lets the analyst assemble, draft, and review in the same place.

In Indago, Search, the Chrome and Brave Data Retriever plug-in, and file uploads all feed one Collection, and drafts are built from the sources the analyst has placed in it. The analyst decides what goes into the Collection.

2. Structured, Repeatable Output
Addresses: rebuilding structure and formatting for every report. Intelligence teams produce the same report types week after week, such as situation reports, threat digests, and executive briefs. When the structure lives in one analyst's memory, every cycle starts over, and formatting, audience framing, confidence language, and citation style get renegotiated before any analysis is written. BCG found lower burnout among workers whose AI use took over repetitive tasks, and rebuilding the same report structure every cycle is repetitive work that a template can take over.

In Indago, a saved template stores the outline (including any citation instructions), the persona, and a default model selection, and can be loaded for any recurring report.

3. Source Attribution, So Verification Happens at the Source
Addresses: the verification burden. When a draft has no trail back to its sources, reviewing it means holding a claim in mind, finding the document that might support it, and reconciling the two, sentence after sentence. That cost applies even when the draft is largely correct. With source-attributed output, the analyst checks each claim against a named source.

In Indago, drafts cite the sources in the analyst's Collection, in a citation style the analyst selects (MLA, Chicago, APA, Harvard, or IEEE) or defines in the outline. A reviewer knows which source to open for each claim. In the sanctions example above, the reviewer opens the named article and checks the date and the wording. Review remains the analyst's job, with a shorter search for where each claim came from.

4. Section-Level Review Checkpoints
Addresses: all-or-nothing review of a full draft. When one section is weak, regenerating the whole report discards the sections that were already good and restarts review on all of them. Revising only the weak section keeps validated work intact and puts attention on the part that needs it. Say a six-section threat brief has a weak executive summary: the analyst rewrites that one section with tighter instructions, and the other five keep the review they have already had.

In Indago, analysts isolate a single section, write targeted instructions, and regenerate only that section. Teammates can comment inline, so feedback stays attached to the passage it refers to.

Tradeoffs Worth Naming Before You Adopt Any Tool

No tool removes the verification burden of AI output entirely, and Indago is no exception. Drafts still need analyst review. Source attribution makes that review faster than checking unattributed output, because a reviewer can trace a claim to a named source, and the analyst still validates sources, applies judgment, and owns the final product.

Each tactic also has an up-front cost. Draft quality follows source quality and outline detail, so a Collection takes careful curation, templates need design time and upkeep as reporting requirements change, and section-level review adds checkpoints to the process.

Partsol reported cutting a due diligence report process from three to four days to one to two, saving roughly $3,800 per report, with the remaining work centered on formatting and flow. That is a result in time and cost, and it does not measure cognitive load.

Four questions to put to any AI tool before it enters an intelligence or security workflow:

• Where does each claim in the output come from?

• Can a reviewer check it without leaving the page?

• Does it add a tool to the stack or remove one?

• What does review cost per report, in time and attention?

Build Review Into the Reporting Workflow

Faster drafting moves analyst time toward review, so reducing cognitive load in AI workflows starts with designing that review deliberately. Book a demo to see how Collections, templates, and section-level review work together in an Indago reporting workflow.