At a security conference, a speaker pitches the AI agent as a “virtual employee”. A sysadmin in the audience described it on a Reddit forum. The thread drew 356 points and 261 comments (exhibit 73).
The replies brought the idea down to earth: an agent is a call to a model inside a loop with tools. Its “memory” is files or a database. Another reply described what that person already sees at work: agents that multiply and become orphans when the person who built them leaves.
The thesis on trial here goes like this. Almost every company already uses AI, agents get better every month and there are success stories in plain sight. So whoever waits falls behind.
If the thesis is true, holding back is expensive. If it is false, rushing is expensive.
To judge it, separate three questions:
Are agents really getting better?
Is using AI the same as winning with it?
Does what works in a pilot make it to production?
A pilot is a small test, with few users and controlled data. Production is real use, with customers, the company’s own systems and a running cost.
The thesis has solid data: agents improve fast and many companies say they see a return.
According to Stanford’s annual AI report (exhibit 75), 88% of organizations use AI.
The same report tracks OSWorld, a benchmark of agent tasks on a computer. A benchmark is a standard test that compares models. On OSWorld, success went from about 12% to about 66%.
METR’s curve points the same way. METR is a lab that measures the time horizon: how long an expert takes on the tasks that AI completes with 50% success. According to its data page (exhibit 55), that horizon has doubled every 128.7 days since 2023, about 4.2 months.
Some measurements sit closer to real work. The Remote Labor Index (exhibit 11) uses 240 freelance projects that a professional had already done. A project counts as automated if the AI delivers something at least as good as the human one.
At launch, in late 2025, the best agent reached 2.5%. In under eight months it climbed to 15.8% (exhibit 115). Today’s public leaderboard has the top agent at 21.25% (exhibit 150).
Results inside companies are improving too. Wharton surveyed about 800 decision makers at large US companies (exhibit 10). Three out of four say they see a positive return from generative AI.
BCG’s 2026 study surveyed 1,330 executives (exhibit 135). About 48.5% of companies already get value from AI.
Menlo Ventures surveyed 495 decision makers (exhibit 32). There, 47% of AI deals reach production, against 25% for traditional software.
AI deals that reach production, versus traditional software.Exhibit 32
Deals that reach production, in %
AI deals47%
Traditional software25%
0100%
Intent keeps pace. Deloitte surveyed 3,235 leaders in 24 countries (exhibit 90). Of them, 74% expect to use agents at least moderately by 2027.
A rising curve does not tell you where things stand today. The Remote Labor Index’s 21.25% leaves about 4 in 5 projects below professional quality. Scale AI, a co-author of the benchmark, runs that leaderboard.
SWE-Bench Pro (exhibit 20), also from Scale AI, tells the same story for software. Its paper compares it with SWE-bench Verified, an easier benchmark. Models that passed 70% on SWE-bench Verified solved less than half of the public set.
On the commercial set, the best one reached 17.8%. With 2026 models, the top of the public leaderboard now solves 61.50% (exhibit 151). The commercial set has no updated leaderboard.
METR’s metric has its own limit: 50% success is a low bar for delegating work. The horizon at 80% success is 4 to 10 times shorter.
Also, in September 2026 METR said its page is no longer updated. Chapter 3 comes back to this.
Agents are improving so fast that they will soon do office work on their own.
Verdict: The improvement is real and fast, but the floor is still low. Since 2023, the length of the tasks they solve with 50% success has doubled every 128.7 days. That is about 4.2 months. On real freelance projects, the best agent went from 2.5% to 21.25% in a year. About 4 in 5 projects still fall short of professional quality.
Rule
Decide based on what the agent does today on your tasks, not on the curve. Rerun the test on your own cases every time you change models.
Why medium confidence. The sources are primary and agree on the trend. But they measure benchmarks (standard tests), mostly in software. The freelance-project table is kept by Scale AI, a coauthor of the benchmark. METR no longer updates its horizons page.
Show the 8 exhibits in detailHide the detail3 for, 2 against, 3 qualify
METR data page with time horizon 1.1. It measures the length of tasks that each model solves with 50% and with 80% success, in the time of an expert person. The trend is still exponential, and since 2023 the horizon doubles faster than in the 2025 paper. But the 80% success horizon is 4 to 10 times shorter than the 50% one. METR warns that above 16 hours its measurement is no longer reliable.
Numbers
128.7 days (interval of 104.4 to 158.0)Time for the 50% horizon to double, counting from 2023 (about 4.2 months). Since 2019 it is 187.8 days. It excludes measurements above 16 hours.
293 minutes (162 to 624) and 49 minutes (21 to 105)Horizon of Claude Opus 4.5 with 50% and with 80% success, with the data corrected on March 3, 2026.
719 minutes (317 to 3,634) and 70 minutesHorizon of Claude Opus 4.6 with 50% and with 80% success.
1,045 minutes (509 to 3,304) and 186 minutesHorizon of Claude Mythos Preview (early version, April 2026) with 50% and with 80% success: about 17.4 hours and 3.1 hours. It is above the 16-hour ceiling that METR considers reliable.
352, 342 and 384 minutes50% horizon of GPT-5.2, GPT-5.4 and Gemini 3.1 Pro.
Limitation The figures come from the page data file (benchmark_results_1_1.yaml). On March 3, 2026 METR corrected a regularization mistake, so 1.1 figures published before that date do not match. On September 8, 2026 METR said the page is no longer actively updated. The tasks are software, machine learning and cybersecurity, and 50% success is a low bar for delegating.
“Measurements above 16 hrs are unreliable with our current task suite.”
Follow-up by the Remote Labor Index team. In less than eight months, the best agent went from automating 2.5% to 15.8% of the freelance projects. They also tried an automated judge calibrated against human evaluation with earlier models. The judge ranked the new models well, but greatly inflated how often they succeed. The authors conclude it works to track relative progress, not to measure absolute capability.
Numbers
15,8 %Automation rate of the best agent (Fable 5). Opus 4.8: 8.3%; GPT-5.5: 6.25%; Gemini 3 Pro: 1.25%.
6.25% versus 17.9%GPT-5.5 according to human evaluators and according to the automated judge (about 2.9 times). For Opus 4.8: 8.33% versus 18.8%.
ρ = 0,90Rank correlation (Spearman) between the judge and human evaluation.
Limitation Published by the benchmark team. A press article gives 16.1% and says only 218 of 240 projects were evaluated. The post does not mention it and the base uses the post figure. Over 84% of the projects still do not reach professional level.
“not a substitute for human evaluation of absolute capability”
An annual compilation of AI data. Adoption in organizations is broad and agents improve fast on benchmarks, but documented AI incidents also rise.
Numbers
88%AI adoption in organizations.
from about 12% to about 66%Success on OSWorld, a benchmark of agent tasks on a computer.
from 60% to almost 100%SWE-bench Verified in one year.
233 and 362Documented AI incidents in 2024 and in 2025.
Limitation It compiles third-party sources (benchmarks, surveys, incident databases). A benchmark is not the same as performance in production. The key findings page has no 2026 inference costs.
Mazeika, M., Gatti, A., et al. arXiv (Center for AI Safety, Scale AI)
This benchmark (a standard test to compare agents) uses 240 real freelance projects instead of a simulated company. A professional already did each project. A project counts as automated only if the AI deliverable is at least as good as the human one, according to trained evaluators. At launch, no agent automated more than 3% of the projects.
Numbers
240 projects in 23 subcategoriesSize of the benchmark: over 6,000 hours of human work and over 140,000 dollars paid; median of 11.5 hours per project.
2,5 %Automation rate of the best agent at launch (Manus). Grok 4 and Claude Sonnet 4.5 reached 2.1%; GPT-5, 1.7%.
94,4 %Agreement between evaluators on the automation rate.
Limitation 2025 models. Evaluation is human and expensive; the authors say it cannot be automated with models today. Co-authored by Scale AI, which sells data and evaluation.
It is a software engineering benchmark that is harder than SWE-bench Verified and less contaminated (fewer problems already seen in training). It uses public repositories, held-out repositories and private commercial repositories. The same models that pass 70% on SWE-bench Verified solve less than half of the public set. On the commercial set they solve less than a fifth.
Numbers
1,865 problems in 41 repositoriesSize of the benchmark.
43.6% and 41.8%Public set: Claude Sonnet 4.5 and GPT-5 (high).
17.8% and 15.7%Commercial set: Claude Opus 4.1 and GPT-5 (high).
Limitation The date is that of version 2 (November 14, 2025), which changes the results: it adds Claude Sonnet 4.5 and Claude Haiku 4.5, released after version 1 (September 21, 2025). The figures in this entry are from version 2. 2025 data: 2026 models are not in the paper. Published by Scale AI, which sells data and evaluation.
Public results table of the Remote Labor Index. In early November 2026, the best agent does about 1 in 5 real freelance projects at professional level. When the benchmark launched, a year earlier, the best reached 2.5%.
Numbers
21,25 %Automation rate of the top entry (Opus 5.5). Next come GPT 6 Astra (20.83%), Fable 5.1 (17.92%) and Gemini Argon 4 (17.08%).
Limitation The page shows no date: the date is the day it was checked (November 2, 2026) and the table can change. It is maintained by Scale AI, co-author of the benchmark. About 4 in 5 projects still do not reach professional level.
Public results table for the public set of SWE-Bench Pro. With 2026 models, the best already solves more than half: about 6 in 10 problems. A year earlier, in the paper, the best reached 43.6%. The private commercial set has no updated table.
Numbers
61.50% (± 3.10)Top entry (Muse Spark 1.1), on 731 tasks.
59.10%, 51.90% and 46.10%gpt-5.4 (xHigh), claude-opus-4-6 (thinking) and gemini-3.1-pro (thinking).
Limitation The page shows no date: the date is the day it was checked (November 2, 2026). It is maintained by Scale AI, the benchmark author. The commercial set page only repeats the paper figures.
METR expanded its task set and re-estimated the time horizon. The horizon is how long a person takes on the tasks that AI solves with 50% success. The overall trend barely changes, but the trend of recent years is faster.
Numbers
228 tasks (previously 170)Size of the task set.
130.8 days and 88.6 daysTime for the horizon to double, measured since 2023 (interval of 107 to 161 days) and since 2024.
293 minutes50% horizon of Claude Opus 4.5 after METR's correction of March 3, 2026, with an interval of 162 to 624 minutes. The January post, written before the correction, gave 320 (170 to 729).
Limitation On March 3, 2026 METR corrected a regularization mistake that changed the measurements. The post still shows the January figures. The Claude Opus 4.5 figure in this entry is the corrected one, taken from METR's time horizons page. 50% success is a low bar for delegating critical work. Confidence intervals are very wide. Of 31 tasks of 8 hours or more, only 5 have a measured human baseline. The tasks are software, machine learning and cybersecurity.
The 88% adoption figure measures use, not return. The return figures are self-reports: what executives say, not a measurement.
In the Wharton survey, optimism drops lower in the organization. Of vice presidents and above, 81% see a positive return, against 69% of middle managers.
The coldest figure is from the IBM study, as reported by Business Today (exhibit 96). The article could not be checked against the full report. About seven in ten CEOs say AI is already transforming their core business.
But only 10% say advanced or agentic AI is what drives their growth the most. Agentic AI is AI that acts through agents. Two years earlier, 49% expected generative AI to be.
Read the surveys with care. BCG and IBM sell AI or consulting, and Menlo is a fund that invests in AI companies.
Even so, the picture has moved. The gap between using and winning is still there, but it is narrowing.
If almost every company already uses AI, it must be paying off.
Verdict: Using is not winning, although the gap is narrowing. AI adoption in organizations reaches 88%. Three in four leaders of large US companies say they see a positive return. Almost half of companies say they get value from AI. But only 10% of CEOs see advanced or agentic AI as the main driver of their growth.
Rule
Before approving an agent, write down which business number it has to move, by how much and by when. Without that number there is no pilot, only a demo.
Why medium confidence. All the return figures are executive self-reports, not measurements. Two of the surveys come from companies that sell AI or consulting. The 10% of CEOs comes from a press article. That figure was not checked against the full report.
Show the 5 exhibits in detailHide the detail1 for, 1 against, 3 qualify
A survey of senior decision makers at large US companies. Three in four say they see a positive return from generative AI, and most already measure that return. Middle managers are less optimistic than vice presidents. About four in ten see a risk of declining skills.
Numbers
3 in 4Leaders who say they see a positive return (81% of vice presidents and above versus 69% of middle managers), in about 800 responses.
72 %Companies that formally measure return.
43 %See a risk of skill loss.
Limitation It is perception, not measurement. Fieldwork ran June 26 to July 11, 2025; the report was published in October 2025. Companies with 1,000 or more employees.
The article reports figures from IBM's 2026 study that the announcement does not include. About seven in ten CEOs say AI is already transforming their core business. But only one in ten says advanced or agentic AI is what most drives their growth. More than half say their main goal with AI is still running pilots and experimenting.
Numbers
10% (two years earlier, 49% expected generative AI to be)CEOs who say advanced or agentic AI is what most drives growth.
53 %Say their main goal with AI is piloting and experimenting.
Limitation A press article about the IBM study; the figures were not checked against the full report. It is CEO opinion, not a return measurement.
An annual compilation of AI data. Adoption in organizations is broad and agents improve fast on benchmarks, but documented AI incidents also rise.
Numbers
88%AI adoption in organizations.
from about 12% to about 66%Success on OSWorld, a benchmark of agent tasks on a computer.
from 60% to almost 100%SWE-bench Verified in one year.
233 and 362Documented AI incidents in 2024 and in 2025.
Limitation It compiles third-party sources (benchmarks, surveys, incident databases). A benchmark is not the same as performance in production. The key findings page has no 2026 inference costs.
The 2026 edition of BCG's study says the gap narrowed: almost half of companies already get value from AI. Many expect to give agents real autonomy, but almost none have the controls to do so today.
Numbers
about 48.5%Companies that generate value: 7.5% leaders (5% in 2025) and 41% scaling; out of 1,330 executives.
42% and 5%Expect to give agents real autonomy by 2030, and have all the controls today.
about 40%Share of AI value they expect from agents by 2030.
Limitation An announcement by a consultancy that sells AI services; self-report, with no detail on how it classifies companies. Automated download now returns 403, so it carries no verbatim quote.
IBM Institute for Business Value, Oxford Economics
The 2026 edition of IBM's CEO study talks about redesigning leadership and adding AI chiefs. It also talks about leaving more operating decisions to AI. It says success depends more on people adopting the tool than on the technology.
Numbers
76% (26% in 2025)Companies with an AI chief, out of 2,000 CEOs in 33 geographies (February to April 2026).
25% today and 48% expected by 2030Operating decisions made by AI with no human involvement.
83 %Say success depends more on people's adoption than on the technology.
Limitation An announcement by a company that sells AI and consulting; an opinion survey. It does not repeat the return figures of the 2025 edition.
Production is not where almost everything falls apart, but it is not a formality either. Menlo’s 47% shows the first part. The IBM article shows the second: 53% of CEOs say their main goal with AI is still piloting and experimenting.
One study gathered 20 cases and 86 practitioners with agents in production (exhibit 29). It found that reliability is still the top challenge. Reliability here means the agent behaves correctly over time, not once.
The sample is small and the authors chose it.
A small thread on a Reddit agents forum describes that jump from the inside (exhibit 101). Agents work in the pilot and degrade in production. Chapter 3 explains why.
If the pilot works, moving to production is a formality.
Verdict: It is not a formality, but it is not where almost everything fails either. In a survey of US decision makers, 47% of AI deals reach production. For traditional software, 25% do. But 53% of CEOs say their main AI goal is still piloting. And teams that already run agents in production name sustained reliability as their main challenge.
Rule
Plan the move to production from day one: integration with the real systems, an owner, monitoring and running cost. If the pilot has no such plan, do not approve it.
Why medium confidence. The figures on reaching production and on piloting are opinion surveys. One is run by a fund that invests in AI and the other was read in the press. The production study has a small sample. The thread about agents that degrade on the way to production is small.
Show the 5 exhibits in detailHide the detail1 for, 2 against, 2 qualify
Tully, T., Redfern, J., Das, D., Xiao, D. Menlo Ventures
Of what companies deploy as AI, only a small part qualifies as a real agent. Companies buy more than they build (76%). Menlo says, without a figure, that bought solutions reach production sooner. About half of AI deals reach production, well above traditional software.
Numbers
16% and 27%Deployments at large companies and at startups that qualify as real agents.
76% boughtShare of AI that was bought in 2025 instead of built (53% in 2024).
47% versus 25%AI deals that reach production, versus traditional software.
37 billion dollarsEnterprise spending on generative AI in 2025 (11.5 billion in 2024).
Limitation A survey of 495 US company decision makers (November 7 to 25, 2025) plus market data. Menlo is a fund that invests in AI companies. It does not say buying works better. It says more is bought and that bought solutions reach production sooner.
“companies spent $37 billion on generative AI in 2025”
The article reports figures from IBM's 2026 study that the announcement does not include. About seven in ten CEOs say AI is already transforming their core business. But only one in ten says advanced or agentic AI is what most drives their growth. More than half say their main goal with AI is still running pilots and experimenting.
Numbers
10% (two years earlier, 49% expected generative AI to be)CEOs who say advanced or agentic AI is what most drives growth.
53 %Say their main goal with AI is piloting and experimenting.
Limitation A press article about the IBM study; the figures were not checked against the full report. It is CEO opinion, not a return measurement.
A thread about agents that work in the pilot and degrade in production. One comment does the math: with 90% success per step, a 5-step workflow fails about 41% of the time. This happens because errors multiply. Another says it is a system problem, and more code alone does not fix it.
Numbers
90% per step, 41% failure over 5 stepsA commenter's math (0.9 to the power of 5 is about 0.59).
Limitation Small thread (6 points and 30 comments in the archive). The math is arithmetic, without field data, and assumes independent errors. It was read in the public Arctic Shift archive because reddit.com blocks automated reading. The quote was compared letter by letter with the original.
“compounding errors is a systematic problem that you cannot code your way out of”
It is a survey of IT and business leaders involved in AI programs. Most expect to use agents soon, but few have a mature governance model to control them. Intention moves faster than control.
Numbers
3,235 leaders in 24 countriesSurvey sample (January 2026 report).
74%Expect to use agents at least moderately by 2027.
21%Have a mature governance model for agentic AI.
Limitation Self-reported intention, not measured use. Other figures that circulate about this report (for example, how many pilots reached production) were only seen in coverage and are not used.
A case study and survey of practitioners with agents in production. Most evaluate with people rather than benchmarks (standard tests), and let the agent take few steps before someone steps in. Reliability, meaning that the agent behaves correctly over time, remains the main challenge. Teams tackle it with system design.
Numbers
20 cases and 86 practitionersBasis of the study: 20 in-depth interviews and a survey of practitioners across 26 domains.
74%Rely mainly on human evaluation.
68%Run at most 10 steps before a person steps in.
70%Prompt off-the-shelf models instead of tuning their weights.
Limitation A small sample chosen by the authors; it does not represent all companies. Preprint.
“Reliability (consistent correct behavior over time) remains the top development challenge”
Klarna’s CEO talked about the company’s agent in the third quarter 2025 results (exhibit 21). He said it does the work of more than 853 people and saved 60 million dollars. Those are the company’s own figures, unaudited.
The same CX Dive article carries the counterweights. In that quarter, customer service and operations cost 50 million dollars, against 42 a year earlier.
Customer service and operations cost in the third quarter of 2025 and a year earlier.Exhibit 21
Cost, in million dollars
Q3 202550 million dollars
A year earlier42 million dollars
0
The company had already started hiring people again in May 2025. An analyst called it an example of a bad deployment.
Klarna is not the rule either. Gartner surveyed 321 customer service leaders, as reported in a press article (exhibit 56). Only 20% cut staff because of AI.
Most keep their people and serve more customers. Gartner also predicts that half of those who cut will hire again by 2027. That is a prediction, not data.
The success stories companies announce show what an agent can achieve.
Verdict: An announcement is not a measurement. Klarna's CEO said its agent does the work of more than 853 people. He also said it saved 60 million dollars. These are the company's own figures, unaudited. In that quarter, its customer service and operations cost rose from 42 to 50 million dollars year over year. In a Gartner survey, only one in five customer service leaders cut staff because of AI.
Rule
Treat every public case as a hypothesis, not as proof. Ask what was measured, who measured it and what happened a year later.
Why medium confidence. Both sources are press. One article carries the company's figures together with their counterweights. The other covers a Gartner survey whose press release could not be opened. No source measures the case independently, which is exactly what the verdict says.
Show the 2 exhibits in detailHide the detail0 for, 1 against, 1 qualifies
In an October 2025 Gartner survey, only one in five customer service leaders cut staff because of AI. Most keep headcount and serve more customers. Gartner predicts that half of those who cut staff will rehire for similar roles, under different job titles.
Numbers
20 %Customer service leaders who cut staff because of AI, out of 321 surveyed.
50 %Of those who cut, the share Gartner predicts will rehire by 2027.
Limitation A press article about a Gartner release (February 3, 2026) that returned 403. The 50% is an analyst prediction, not data.
“AI simply isn’t mature enough to fully replace the expertise, empathy, and judgment”
At the third quarter 2025 earnings presentation, Klarna's CEO gave two figures. He said its AI agent does the work of more than 853 full-time people and saved 60 million dollars. The same article gives data against this: customer service and operations cost rose versus the prior year. Also, the company went back to hiring people in May 2025, and an analyst calls it an example of bad deployment.
Numbers
over 853 people and 60 million dollarsEquivalent work and savings according to the CEO (it was 700 people in early 2025).
50 versus 42 million dollarsCustomer service and operations cost in the third quarter of 2025 and a year earlier.
Limitation The savings figures are the company's own and unaudited. The article was read; automated download now returns 403, so it carries no verbatim quote.
Agents are improving fast, and the gap between using AI and getting value from it is narrowing.But the return still rests on what executives say. And the capability floor is still low on real work.
Four things sit between the promise and the result: the number the project has to move, the owner, the controls and reliability over time in production. The model is one part of the math.
Control moves slower than intent. In the Deloitte survey, 74% expect to use agents by 2027.
But only 21% have a mature governance model for them. Governance here means knowing who decides, who oversees and who shuts an agent off.
21%
Have a mature governance model for agentic AI. Exhibit 90
In BCG’s survey, 42% expect to give agents real autonomy by 2030. Only 5% have all the controls in place today.
According to IBM (exhibit 95), AI already makes 25% of operational decisions with no human involved. That study surveyed 2,000 CEOs.
Cost does not wait either. A CX Today article reports a Gartner prediction (exhibit 51). By 2030, resolving a customer service case with generative AI will cost more than 3 dollars.
That would be more than many offshore human agents, partly because organizations consume more than they expect. The orphaned agents from the opening are another cost: the cost of adding without an owner.
Rushing without control is expensive, and waiting without doing anything does not help either. The pace is set by what you can govern.
You have to adopt agents as soon as possible: whoever waits falls behind.
Verdict: Rushing without control is expensive. 74% of the leaders surveyed expect to use agents at least moderately by 2027. But only 21% have a mature governance model for them. In another survey, 42% expect to give them real autonomy by 2030. Today, only 5% have all the controls. Gartner predicts that by 2030, a customer service case with AI will cost more than with many outsourced human operators.
Rule
Before adding agents, put the controls in place: an inventory, an owner, a spending cap and a way to switch them off. Move at the pace at which you can govern what you already have.
Why medium confidence. The surveys are large, but they measure intent and self-reports. Consultancies or companies that sell AI run them. The cost point is an analyst prediction read in the press. The point that agents are left orphaned when their creator leaves comes from a Reddit thread.
Show the 5 exhibits in detailHide the detail1 for, 2 against, 2 qualify
IBM Institute for Business Value, Oxford Economics
The 2026 edition of IBM's CEO study talks about redesigning leadership and adding AI chiefs. It also talks about leaving more operating decisions to AI. It says success depends more on people adopting the tool than on the technology.
Numbers
76% (26% in 2025)Companies with an AI chief, out of 2,000 CEOs in 33 geographies (February to April 2026).
25% today and 48% expected by 2030Operating decisions made by AI with no human involvement.
83 %Say success depends more on people's adoption than on the technology.
Limitation An announcement by a company that sells AI and consulting; an opinion survey. It does not repeat the return figures of the 2025 edition.
Gartner predicts that in customer service, the cost per case resolved with generative AI will exceed that of many outsourced human agents. One reason is that consumption is higher than expected and today's prices are subsidized. Also, new models spend more tokens per interaction, and infrastructure costs money.
Numbers
over 3 dollarsCost per resolution with generative AI that Gartner predicts for 2030.
3, 5 or up to 10 timesTokens frontier models use for similar interactions, according to the article.
Limitation A press article about a Gartner release (January 26, 2026) that returned 403. It is a prediction, not measured data.
“many organizations consume a lot more than they expect”
Reporting from a security conference, the author says agents are being sold there as “virtual employees”. The comments bring it down to earth. An agent is a model call in a loop with tools, and its “memory” is files or a database. Another warns that afterwards there is no way to analyze why it decided something. A third describes the sprawl of agents that are orphaned when the person who created them leaves.
Numbers
356 points and 261 commentsSize of the thread according to the archive. The quoted comment has 86 points.
Limitation Opinions from people who run IT, without measurement. It was read in the public Arctic Shift archive because reddit.com blocks automated reading. The quote was compared letter by letter with the original.
It is a survey of IT and business leaders involved in AI programs. Most expect to use agents soon, but few have a mature governance model to control them. Intention moves faster than control.
Numbers
3,235 leaders in 24 countriesSurvey sample (January 2026 report).
74%Expect to use agents at least moderately by 2027.
21%Have a mature governance model for agentic AI.
Limitation Self-reported intention, not measured use. Other figures that circulate about this report (for example, how many pilots reached production) were only seen in coverage and are not used.
The 2026 edition of BCG's study says the gap narrowed: almost half of companies already get value from AI. Many expect to give agents real autonomy, but almost none have the controls to do so today.
Numbers
about 48.5%Companies that generate value: 7.5% leaders (5% in 2025) and 41% scaling; out of 1,330 executives.
42% and 5%Expect to give agents real autonomy by 2030, and have all the controls today.
about 40%Share of AI value they expect from agents by 2030.
Limitation An announcement by a consultancy that sells AI services; self-report, with no detail on how it classifies companies. Automated download now returns 403, so it carries no verbatim quote.
An agent gets approved for the business number it moves, not for the technology it uses. In practice that comes down to four decisions:
Write down the number before approving: which number, by how much and by when. Without that number there is no pilot, only a demo.
Plan the move to production from day one: integration with the real systems, an owner, monitoring and running cost.
Put the controls in place before adding agents: an inventory, an owner, a spending cap and a way to shut them off. Move at the pace at which you can govern what you already have.
Decide based on what the agent does today on your tasks, not on the curve. Repeat the evaluation on your own cases every time you change models.
Now:
Pick the most advanced agent proposal in your organization.
Ask for one page with the number it has to move, the owner and who takes over if they leave.
Add how much it can spend per month and how it gets shut off.
If nobody can fill in that page, you know where the project stands.