Organizations: The University of Texas at Dallas · University of Southern California · Carnegie Mellon University · Georgia Institute of Technology · University of California, Los Angeles · University of Pennsylvania · Amherst College · Vanderbilt University · University of California, Irvine · University of Chicago · Harvard University · Brown University · Texas A&M University · University of Illinois Urbana-Champaign · Purdue University · The Ohio State University · Massachusetts Institute of Technology
Computer-use agents (CUAs) are commonly evaluated against fixed task specifications, but users may omit requirements, add constraints, or require approval before consequential actions. Completing the benchmark instruction does not guarantee serving the person who asked. We introduce PersonaCUABench, which holds 74 tasks across 11 domains fixed and varies only who asks. A persona agent, playing one of 44 personas in four groups, rewrites each instruction in that user's own words and answers the CUA's questions, while the CUA sees only the dialogue, never the profile or the original instruction. Across nine backbones (3,330 runs), mean task completion falls by 36% relative to the original requests. The losses have two sources: in 59.9% of cells that lose points, the agent engages with what the user asked for and still gets it wrong, and 11.2% are stops the user requested that the checker scores as failures. Even when a persona request is fully completed, the user gives the top satisfaction rating in only 62.7% of cells, against 84.1% for the original requests. CUAs should therefore be evaluated not only on task completion but also on whether they meet the requirements and constraints users express. Code: https://github.com/MatrAIx-ai/PersonaCUABench. Project page: https://matraix.ai/research/persona-cua.html
Figures & tables
Figure 1: Overview of PersonaCUABench. Top: 74 tasks in 11 domains, 44 personas, four groups in each of 11 domains, plus the original request, task completion of nine backbones on original and persona requests, and the distribution of user satisfaction. Stage 1: a MatrAIx persona agent with a 37-attribute profile rephrases the original task in the user’s own words. The CUA sees neither the profile nor the task. Stage 2: the CUA sees only the dialogue, acts in Chrome, and may ask the persona agent questions. Stage 3: each rollout gets two scores: task completion against the original task, and satisfaction rated by the same persona agent. Here MiniMax M3 stops before sending the offer, so the task fails while the user rates it 5/5. Each requirement is then checked as raised, engaged, and met to locate the loss.
Odysseys
REAL
Total
Tasks
38
36
74
Task type
Information
Action
—
Environment
Live web
Replicas
—
Requirements
206
107
313
Completion check
Rubric judge
State checker
—
Life / work tasks
28 / 10
12 / 24
40 / 34
Table 1: Task composition. Three domains (travel, food, lifestyle) contain tasks from both sources.
Table 2: The four persona groups K , each instantiated in all 11 domains, with their values on the five grouping attributes. All 11 personas of a group share these values, with five exceptions listed with the attributes.
Model
Task Completion
Satisfaction (1–5)
Original
expert
senior
young
practical
Original
expert
senior
young
practical
MiniMax M3
0.686
0.447
0.334
0.355
0.386
4.01
3.08
3.05
3.09
3.09
Kimi K3
0.768
0.555
0.432
0.452
0.496
4.24
3.88
3.22
3.57
3.47
Muse Glimmer 30B †
0.548
0.413
0.312
0.374
0.316
3.59
3.31
2.88
3.16
2.85
Grok 4.6
0.684
0.543
0.476
0.443
0.515
4.16
3.88
3.14
3.62
3.53
Qwen 3.8 27B †
0.729
0.554
0.431
0.497
0.501
4.09
3.77
3.38
3.66
3.26
Table 3: Task completion and user satisfaction at delivery over all 74 tasks. Models are listed in order of release date, and bold marks the best model in each column. † denotes open-source models.
Table 4: Task completion by scenario and persona group, mean over nine models. Table 14 gives the intervals.
Figure 2: Change in task completion by domain, mean over nine models with 95% intervals.
Figure 3: Task completion and satisfaction at delivery by domain, for the original request and each persona group, mean over nine models. Mixed domains list their information and action tasks.
Figure 4: A practical user’s case, run by DeepSeek V4.1 Flash. Appendix D gives one case for each of the other three groups.
Figure 5: Primary cause of each loss cell, nine models pooled. Each cell has one cause. (a) All 1,230 loss cells. (b) Each persona group, with its number of loss cells.
Benchmark
Computer use
Same task, many users
Can ask the user
Judged by user
WebArena
✓
✗
✗
✗
OSWorld 2.0
✓
✗
✓
✗
MyPCBench
✓
✗
✗
✗
PersonalWAB
✗
✗
✓
✗
Persona2Web
✓
✗
✗
✗
PrefDisco
✗
✓
✓
✗
Table 5: Comparison with related benchmarks.
Appendix figures & tables28 assets
Supplementary material from the paper’s appendix.
Appendix
Who they are
Can they use the result
What they ask for and expect
Which trade-offs they accept
age bracket highest education household income time pressure seniority technology savviness region urbanicity socioeconomic band geographic mobility
interest in technology attitude to new technology device context mobility vision assistive technology accessibility needs
detail orientation patience attention span big picture vs. detail brief vs. full detail planning horizon planned vs. spontaneous
budgeting skill frugality economic motivation budget tracking save vs. spend preference risk tolerance financial risk tolerance (DOSPERT) risk framing
Appendix
Table 6: The 32 general attributes shared by all personas, grouped by the question they answer (10, 7, 7, and 8 attributes).
Domain
Industry
Skills
Knowledge
Interests
Other
Art
fine art
drawing design thinking
art history
–
appreciation of beauty
Community
nonprofit government
facilitation
–
volunteering
value: community
E-commerce
e-commerce retail
negotiation
–
–
shopping style brand loyalty
Food
food and beverage hospitality
cooking
–
cooking
diet type
Health
healthcare fitness and wellness
–
nutrition
fitness
exercise frequency
Lifestyle
apparel real estate
–
–
interior design fashion
fashion sense
Appendix
Table 7: The five domain-specific attributes of each domain, grouped by type. Knowledge covers familiarity with a field and subject knowledge; Other covers values, styles, and habits. A dash marks a type that a domain does not use.
Figure 13
Figure 6: How often each profile dimension appears in the persona’s utterances (2,664 persona cells over nine models). The right column gives the share of cells in which the dimension is clearly expressed.
Model
Clear (%)
Contradiction (%)
Δ : all → excluded
Accessibility (%)
Device (%)
MiniMax M3
16.1
11.1
−0.305→−0.317
1.7
2.0
Kimi K3
14.5
9.1
−0.284→−0.285
2.4
1.4
Muse Glimmer 30B
15.0
11.8
−0.194→−0.191
3.4
4.7
Grok 4.6
13.8
8.8
−0.190→−0.194
3.7
2.7
Qwen 3.8 27B
15.0
12.5
−0.233→−0.225
3.4
4.4
Gemini 3.8 Flash
13.5
7.8
−0.311→−0.313
2.0
1.7
Appendix
Table 8: Persona fidelity per model. Clear is the share of the 11 profile dimensions that the persona’s utterances state clearly. Contradiction is the share of cells with at least one utterance that contradicts the profile. The Δ column gives the overall Δ on all cells and with those cells excluded. Accessibility and Device give the share of cells in which the accessibility or device dimension is clearly expressed.
εi=1
εi=0
ρi=1
engaged
dropped
ρi=0
engaged unprompted
unsaid
Appendix
Table 9: Intent outcomes for a requirement ri , by whether the user raised it ( ρi ) and whether the agent engaged with it ( εi ).
Model
List price, input / output (USD per million)
Input tokens (millions)
Output tokens (millions)
Mean steps
Agent cost (USD)
Agent cost per cell (USD)
Persona agent and judges, estimated (USD)
MiniMax M3
0.30 / 1.20
332.1
3.39
43.1
104
0.28
18
Kimi K3
3.00 / 15.00
372.9
5.75
35.8
1,205
3.26
17
Muse Glimmer 30B
–
293.6
2.49
41.6
–
–
25
Grok 4.6
2.00 / 6.00
397.4
2.55
42.9
810
2.19
26
Qwen 3.8 27B
0.42 / 3.00
404.0
5.69
35.1
187
0.50
25
Gemini 3.8 Flash
0.75 / 3.75
384.6
3.02
36.8
300
0.82
21
Appendix
Table 10: API cost per backbone at list prices without cache discounts. The persona agents and judges run on GPT-5.6-terra and are estimated at 2.00 / 12.00 USD per million input / output tokens. † Eight backbones with a list price.
Judgment
Items
P1
P2
Requirement
Met
3,957
86.2 (0.72)
88.2 (0.75)
Raised
3,645
88.1 (0.47)
84.7 (0.20)
Engaged
3,645
97.2 (0.55)
95.9 (0.30)
Profile dimension
Clear
8,712
90.7 (0.59)
88.1 (0.54)
Appendix
Table 11: Agreement of human annotators P1 and P2 with the GPT-5.6-terra judge on 22 representative tasks (two per domain), as percent agreement with Cohen’s κ in parentheses. Items count the labels that GPT-5.6-terra gives. For each profile dimension, Clear asks whether it is clearly expressed, and Level asks whether it is clearly, weakly, or not expressed.
Compared with
Pairs
P1
P2
Persona agent’s own rating
At delivery
987
37.5 ( 0.37 )
37.3 ( 0.44 )
After verification
305
32.1 ( 0.33 )
36.4 ( 0.30 )
External judge (P5)
At delivery
989
48.6 ( 0.52 )
58.2 ( 0.60 )
Appendix
Table 12: Agreement of human annotators P1 and P2 with the two automated satisfaction ratings on the 22 selected tasks, as exact percent agreement with quadratic-weighted Cohen’s κ in parentheses. Pairs are matched on backbone, task, persona, and rating stage.
Model
Δ
95% CI
Relative (%)
Δ excluded
MiniMax M3
−0.305
[ −0.379 , −0.229 ]
−45
−0.317
Kimi K3
−0.284
[ −0.358 , −0.213 ]
−37
−0.285
Muse Glimmer 30B
−0.194
[ −0.276 , −0.119 ]
−35
−0.191
Grok 4.6
−0.190
[ −0.256 , −0.124 ]
−28
−0.194
Qwen 3.8 27B
−0.233
[ −0.303 , −0.169 ]
−32
−0.225
Gemini 3.8 Flash
−0.311
[ −0.381 , −0.240 ]
−42
−0.313
Appendix
Table 13: Δ , 95% interval, relative change, and Δ after exclusion, for Table 3 . Δ is task completion on the persona request minus that on the original request on the same task, averaged over the four groups. Relative is Δ divided by the original-request score, in percent. Δ excluded drops the 11.1% of cells in which the persona agent contradicts its profile. Every interval lies below zero, and excluding those cells changes the mean Δ from −0.248 to −0.249 . Δ pairs each persona cell with the original-request cell of the same task.
Figure 7: Task completion and satisfaction at delivery per model, for the original request and each persona group, over the 74 tasks. The gray bar spans the four persona groups, and the black marker is the original request. Models are sorted by mean persona task completion.
Group
Tasks
Original → persona
Change [95% CI]
Relative change (%)
All
74
0.689 → 0.441
−0.248 [ −0.295 , −0.202 ]
−36
Work
34
0.705 → 0.431
−0.274 [ −0.352 , −0.197 ]
−39
Life
40
0.675 → 0.449
−0.226 [ −0.280 , −0.174 ]
−33
expert
74
0.689 → 0.504
−0.185 [ −0.240 , −0.135 ]
−27
senior
74
0.689 → 0.402
−0.287 [ −0.343 , −0.232 ]
−42
young
74
0.689 → 0.414
−0.275 [ −0.326 , −0.225 ]
−40
Appendix
Table 14: Task completion by scenario and persona group, all tasks pooled, mean over nine models. Change: persona minus original on the same task, with a task-level bootstrap interval. In each of the seven groups, all nine models lose.
Domain
User
Mean
Professional (14)
Original
0.857
0.929
0.714
0.786
0.857
0.786
0.786
0.643
0.714
0.786
expert
0.357
0.429
0.214
0.571
0.357
0.286
0.357
0.214
0.429
0.357
senior
0.357
0.571
0.357
0.714
0.500
0.429
0.286
0.429
0.429
0.452
young
0.500
0.500
0.429
0.500
0.571
0.429
0.286
0.214
0.500
0.437
practical
0.357
0.500
0.214
0.643
0.571
0.429
0.286
0.286
0.286
0.397
Productivity (9)
Original
0.444
0.667
0.333
0.444
0.444
0.556
0.333
0.556
0.556
0.481
Appendix
Table 15: Task completion by domain, user group, and model. Mixed domains list their information and action tasks. Model columns follow the logos in order of release, as in Table 3 : MiniMax M3, Kimi K3, Muse Glimmer 30B, Grok 4.6, Qwen 3.8 27B, Gemini 3.8 Flash, GPT-6 Astra, DeepSeek V4.1 Flash, and Claude Opus 5.5. Bold: best model in each row.
Domain
User
Mean
Professional (14)
Original
4.79
5.00
4.50
5.00
5.00
5.00
4.79
4.71
4.86
4.85
expert
3.57
3.93
3.86
4.50
4.07
3.57
4.21
4.14
4.07
3.99
senior
3.71
4.07
3.21
3.93
4.21
3.43
3.29
4.00
4.43
3.81
young
4.21
4.93
4.14
4.71
4.57
4.21
4.36
4.57
4.21
4.44
practical
4.14
4.14
3.00
3.93
3.43
3.64
4.00
4.29
4.14
3.86
Productivity (9)
Original
4.67
4.67
4.78
4.56
4.11
5.00
4.33
4.00
4.33
4.49
Appendix
Table 16: Satisfaction at delivery by domain, user group, and model. The original request is rated by the same model without a profile. Model columns match Table 15 . Bold: best model in each row.
Figure 8: Task completion by domain for the original request and each persona group, mean over nine models. The original request is highest in every domain. Table 15 gives the values for each model.
Figure 9: Satisfaction at delivery by domain for the original request and each persona group, mean over nine models. The original request is rated by the same model without a profile. Table 16 gives the values for each model.
Model
Action AUC
Failed, judged done (%)
Information correlation
Misreport (%)
External judge
Delivery
Verified
Delivery
Verified
External judge
Delivery
Information
Action
MiniMax M3
0.56
0.66
0.76
54
33
0.28
0.51
58
38
Kimi K3
0.58
0.62
0.78
62
28
0.29
0.54
49
22
Muse Glimmer 30B
0.66
0.73
0.81
42
25
0.46
0.63
60
33
Grok 4.6
0.49
0.55
0.69
64
38
0.34
0.62
36
17
Qwen 3.8 27B
0.56
0.63
0.75
58
30
0.27
0.57
43
32
Appendix
Table 17: How well each rating tracks task completion, on persona cells. AUC measures separation of passed from failed action cells (0.5 is chance). Information correlation is the correlation between the rating and the information-task score. At delivery the user sees the final reply and the open tabs, and when verified the user also sees the resulting site state. Verified ratings cover only a subset of the tasks, namely those that change the site. Misreport is the share of loss cells whose final reply is inconsistent with the outcome.
Group
Cells
At delivery
Verified
Rating
Done (%)
Rating
Done (%)
Stopped correctly
138
3.69
62
1.86
7
Other failures
205
3.25
46
2.42
27
reply misreports
75
3.47
55
2.63
33
reply is accurate
130
3.13
41
2.31
24
Appendix
Table 18: Satisfaction rating from 1 to 5 and share of cells the user judges done, for the 343 action loss cells. Verified ratings are given after the user checks the site and cover only the tasks that change a site.
Table 19: Satisfaction from 1 to 5 on each group’s own successful and failed cells, pooled over nine models and all tasks. Yes and No mark whether the cell is fully successful. After-verification ratings cover only the tasks that change a site.
Label (% of loss cells)
Information
Action
Missed a raised requirement a
76.7
24.2
Missed a requirement not raised a
45.7
1.7
only requirements not raised missed a
23.3
0.9
Final step not executed b
–
74.9
of which a correct stop
–
40.2
User asked for less or for staged execution
14.7
59.2
Appendix
Table 20: Labels of the 887 information and 343 action loss cells. A cell can carry several labels. The final step is to submit, book, or send. Labels marked a come from the per-requirement scores, those marked b from the final site state, and the rest from GPT-5.6-terra.
Figure 10: Share of fully successful cells and mean steps of successful and failed cells, by user group, pooled over nine models. Only 21 information cells are fully successful for senior and 22 for young .
Model
Asked a question (%)
Misreport (%)
Judged done (%)
Information
Action
Information
Action
MiniMax M3
4.6
40.3
58
38
54
Kimi K3
4.6
22.9
49
22
62
Muse Glimmer 30B
3.3
16.7
60
33
42
Grok 4.6
1.3
2.8
36
17
64
Qwen 3.8 27B
3.3
27.1
43
32
58
Appendix
Table 21: Clarification and outcome reporting by model, in the order of Table 3 . Asked a question counts persona cells, misreport counts loss cells, and judged done counts failed action cells. † Pooled over cells.
Model
Ask, information (%)
Ask, action (%)
Questions
On target (%)
Answers (%)
Asked − not asked Difference (tasks)
Persona
Original
Persona
Original
Invented
Conflict
MiniMax M3
4.6
5.3
40.3
25.0
96
21
73
3
+0.103 (26)
Kimi K3
4.6
2.6
22.9
8.3
55
36
73
11
+0.164 (16)
Muse Glimmer 30B
3.3
5.3
16.7
8.3
31
10
97
0
−0.185 (9)
Grok 4.6
1.3
0.0
2.8
0.0
9
33
89
22
−0.500 (5)
Qwen 3.8 27B
3.3
0.0
27.1
13.9
64
23
72
9
+0.074 (16)
Appendix
Table 22: Clarification behavior. Ask is the share of cells with at least one question. On target counts questions that target a requirement the user omitted. Invented counts persona-agent answers that add details absent from the task, and Conflict counts answers whose details conflict with the task. Difference is, within a task, mean task completion of persona cells in which the agent asked minus that of cells in which it did not, with the number of tasks containing both kinds of cell.
Group
Information tasks
Action tasks
Δ completion [95% CI]
Δ satisfaction [95% CI]
Median length (chars)
Δ completion [95% CI]
Better / worse
expert
−0.028 [ −0.065 , +0.009 ]
+0.34 [ +0.16 , +0.53 ]
4148 → 4479
+0.083 [ −0.028 , +0.194 ]
4 / 1
senior
−0.070 [ −0.115 , −0.025 ]
−0.18 [ −0.42 , +0.05 ]
2688 → 2378
0.000
0 / 0
young
−0.015 [ −0.068 , +0.038 ]
−0.05 [ −0.24 , +0.13 ]
2750 → 2506
+0.083 [ 0.000 , +0.194 ]
3 / 0
practical
+0.004 [ −0.052 , +0.065 ]
−0.13 [ −0.32 , +0.05 ]
3952 → 3520
−0.056 [ −0.139 , 0.000 ]
0 / 2
All
−0.027 [ −0.051 , −0.001 ]
−0.01 [ −0.09 , +0.09 ]
3469 → 3185
+0.028 [ −0.007 , +0.069 ]
7 / 3
Appendix
Table 23: Effect of showing the agent the user’s profile (GPT-6 Astra, 296 cells with identical opening requests). Each change is the profile-visible value minus the main-experiment value on the same cell. Satisfaction is the user’s rating at delivery. Better / worse counts the action cells whose outcome changes. Bold: the interval excludes zero.
Slice
Cells
Task completion: main → forced
Δ [95% CI]
Better / worse
Asked (%)
Steps
Information, original
38
0.762 → 0.776
+0.013 [ −0.030 , +0.054 ]
8 / 5
21 → 100
41.1 → 49.1
Information, persona
152
0.582 → 0.549
−0.032 [ −0.065 , +0.001 ]
28 / 42
23 → 100
37.0 → 50.0
Action, original
36
0.583 → 0.472
−0.111 [ −0.250 , 0.000 ]
1 / 5
25 → 100
17.0 → 25.7
Action, persona
144
0.278 → 0.285
+0.007 [ −0.042 , +0.056 ]
8 / 7
53 → 100
18.8 → 24.5
All persona
296
0.434 → 0.421
−0.013 [ −0.042 , +0.018 ]
36 / 49
38 → 100
28.1 → 37.6
Appendix
Table 24: Effect of forcing the agent to ask at least one clarification question (GPT-6 Astra, 370 cells with identical opening requests). Δ is the forced value minus the main-experiment value on the same cell. Better / worse counts the cells whose score rises or falls, Asked is the share of cells with at least one question, and Steps is the mean number of agent steps.
Figure 11: Senior persona case in PersonaCUABench, run by DeepSeek V4.1 Flash. A cost-cautious user requests a fixed price before the agent sends an offer. The sampled views show job review, candidate review and an unsubmitted offer form. The agent returns a $1,800 draft and states that nothing was sent. The original checker awards 0/1 for the unfinished offer, although the pause follows the boundary the user added. The user, who still wanted the hire, rates it 2/5. Note that, Figure 1 is run by MiniMax, with the same senior user which also stops before sending the offer and is rated 5/5.
Figure 12: Expert persona case in PersonaCUABench run by DeepSeek V4.1 Flash. The user requests reusable sources and a cited workshop brief. The agent opens the three evidence pages and the album page, then supplies the sources and track titles. The saved rubric judge awards 5/7. R4 and R6 fail because the final synthesis and song rationale do not meaningfully integrate the evolutionary-bias evidence, despite opening its source.
Figure 13: Young persona case in PersonaCUABench run by DeepSeek V4.1 Flash. The agent searches for a low-cost group stay, opens the cheaper listing and explores messaging. The user agrees to forward the link, and the final answer states that the agent did not send it. The saved checker awards 1/1, but its implementation checks search criteria only. This case exposes a gap between the checker’s pass and the original task’s group-chat delivery requirement.
Current benchmarks for computer-use agents evaluate models in impersonal environments. This leaves a gap between evaluation and deployment where personal assistants are expected to work across a user's whole digital life, including their context, historical data, and logged-in accounts. This gap is widest on web tasks, where live web evaluations cannot exercise sites that require logging in or personal information, the kind of site a real personal assistant has to drive. We introduce MyPCBench, which tests computer-use agents as personal assistants on a Linux desktop populated with 17 simulated real-world web applications and a full desktop stack, all seeded for one canonical persona, Michael Scott from The Office. We define 184 tasks in this environment, each inspired by a real request drawn from the OpenClaw community, and benchmark six closed and open-weight models with a uniform computer+bash tool surface. We find that the best model, Claude Opus 4.6, fully solves 55.4% of the tasks, the only model above 50%. Model failures cluster on tasks that span many applications and on long trajectories, where personalization stresses an assistant the most. We release the environment, task set, and agent harness at https://mypcbench.com.
Lawrence Keunho Jang, Andrew Keunwoo Jang, Jing Yu Koh +1
We introduce AgentPersonaBench (APB), a benchmark evaluating whether persona conditioning faithfully steers downstream agent behavior. While language models are increasingly deployed for persona-driven user simulation, existing benchmarks primarily evaluate conversational styling or self-reports rather than authentic behavioral fidelity. APB evaluates latent persona adherence one trait at a time, embedding each target trait within a complete synthetic profile without explicitly naming the trait or disclosing the test. Ground-truth adherence is verified strictly from observable actions across four interaction surfaces of increasing realism: survey, chat, web (interactive web environments), and app (desktop software environments). APB comprises 2,460 tasks spanning 867 traits, verified through automated audits and expert review. Our evaluation of 20 frontier model arms demonstrates that high-fidelity user simulation is already attainable: leading models achieve up to 84.7% full-pass adherence under unprompted conditions. At the same time, APB identifies clear behavioral boundaries: adherence drops across interaction modalities (only 37.9-64.3% pass all four surfaces), multi-attribute demands degrade retention, and competing model families exhibit pronounced behavioral divergence.
Personalized mobile agents that infer user preferences and calibrate proactive assistance hold great promise as everyday digital assistants, yet existing benchmarks fail to capture what this requires. Prior work evaluates preference recovery from static histories or intent prediction from fixed contexts. Neither tests whether an agent can elicit missing preferences through interaction, nor whether it can decide when to intervene, seek consent, or remain silent in a live GUI environment. We introduce KnowU-Bench, an online benchmark for personalized mobile agents built on a reproducible Android emulation environment, covering 42 general GUI tasks, 86 personalized tasks, and 64 proactive tasks. Unlike prior work that treats user preferences as static context, KnowU-Bench hides the user profile from the agent and exposes only behavioral logs, forcing genuine preference inference rather than context lookup. To support multi-turn preference elicitation, it instantiates an LLM-driven user simulator grounded in structured profiles, enabling realistic clarification dialogues and proactive consent handling. Beyond personalization, KnowU-Bench provides comprehensive evaluation of the complete proactive decision chain, including grounded GUI execution, consent negotiation, and post-rejection restraint, evaluated through a hybrid protocol combining rule-based verification with LLM-as-a-Judge scoring. Our experiments reveal a striking degradation: agents that excel at explicit task execution fall below 50% under vague instructions requiring user preference inference or intervention calibration, even for frontier models like Claude Sonnet 4.6. The core bottlenecks are not GUI navigation but preference acquisition and intervention calibration, exposing a fundamental gap between competent interface operation and trustworthy personal assistance.