Coded turns from 429 human-AI pair-programming ("vibe coding") sessions
across 34 projects, held between 5 February and 8 March 2026, in long
format with one row per code. human_long holds the codes of the human
turns and ai_long those of the AI turns (428 of the sessions have AI
turns). The two data frames share their columns and their sessions, and
interleave on order_in_session.
Format
Two data frames with the same 9 columns:
- message_id
Integer identifier of the message (turn); the codes of one message share it, and no identifier occurs in both data frames.
- project
Character project identifier,
Project_1toProject_34.- session_id
Character session identifier, a 12-character hash.
- timestamp
Integer Unix time (seconds) of midnight, UTC+2, on
session_date. It has day resolution, so the rows of one day share it;build_network(time = "timestamp")keeps tied rows in row order, which followsorder_in_session. With the defaulttime_threshold(900 seconds), each change of day also starts a new sequence, so the 84 sessions that run over more than one day are split (526 sequences instead of 429); to keep every session whole, usetime = "order_in_session".- session_date
Character date of the row (
YYYY-MM-DD); 84 sessions run over more than one day.- code
Character interaction code. The two data frames use disjoint code sets: 9 labels in
human_long(Command, Correct, Frustrate, Inquire, Interrupt, Refine, Request, Specify, Verify) and 8 inai_long(Ask, Delegate, Execute, Explain, Investigate, Plan, Repair, Report), 17 in total.- cluster
Character high-level cluster of the code.
human_longuses Directive, Evaluative and Metacognitive;ai_longuses Action, Communication and Repair.- code_order
Integer order of the code within its message, starting at 1.
- order_in_session
Integer position of the row in its session, counting the rows of both data frames together.
human_long | Human turns only, 10,796 rows |
ai_long | AI turns only, 8,551 rows |
An object of class data.frame with 10796 rows and 9 columns.
An object of class data.frame with 8551 rows and 9 columns.
Source
Saqr, M. (2026). Human-AI vibe coding interaction study. https://saqr.me/blog/2026/human-ai-interaction-cograph/
Examples
net <- build_network(human_long, method = "tna",
action = "code", actor = "session_id",
time = "timestamp")
#> Metadata aggregated per session: ties resolved by first occurrence in 'cluster' (20 sessions)
net
#> Transition Network (relative probabilities) [directed]
#> Weights: [0.018, 0.620] | mean: 0.111
#>
#> Weight matrix:
#> Command Correct Frustrate Inquire Interrupt Refine Request Specify
#> Command 0.235 0.088 0.055 0.066 0.035 0.038 0.155 0.280
#> Correct 0.093 0.091 0.138 0.057 0.054 0.112 0.120 0.278
#> Frustrate 0.103 0.115 0.176 0.075 0.047 0.171 0.111 0.125
#> Inquire 0.196 0.126 0.098 0.188 0.094 0.062 0.084 0.106
#> Interrupt 0.259 0.081 0.094 0.123 0.102 0.080 0.069 0.122
#> Refine 0.058 0.075 0.072 0.047 0.042 0.086 0.146 0.457
#> Request 0.096 0.019 0.044 0.067 0.051 0.033 0.039 0.620
#> Specify 0.263 0.058 0.081 0.072 0.167 0.074 0.070 0.172
#> Verify 0.224 0.079 0.164 0.118 0.043 0.097 0.116 0.083
#> Verify
#> Command 0.048
#> Correct 0.056
#> Frustrate 0.077
#> Inquire 0.044
#> Interrupt 0.069
#> Refine 0.018
#> Request 0.032
#> Specify 0.043
#> Verify 0.077
#>
#> Initial probabilities:
#> Specify 0.679 ████████████████████████████████████████
#> Command 0.175 ██████████
#> Request 0.055 ███
#> Interrupt 0.030 ██
#> Correct 0.019 █
#> Refine 0.019 █
#> Frustrate 0.013 █
#> Inquire 0.008
#> Verify 0.002