Human-AI Vibe Coding Interaction Data

Coded interaction sequences from 429 human-AI pair programming sessions across 34 projects. Three coding granularities: code (32 states), category (17 states), and superclass (6 states).

Usage

human_ai

human_ai_cat

human_ai_super

human_detailed

human_cat

human_super

ai_detailed

ai_cat

ai_super

human_wide

ai_wide

Format

Long-format data frames with columns:

id

Integer. Turn index within the session.

project

Character. Project identifier (Project_1 .. Project_34).

session_id

Character. Unique session hash.

timestamp

Character. ISO 8601 timestamp.

session_date

Character. Date of the session (YYYY-MM-DD).

actor

Character. "Human" or "AI".

code

Character. Fine-grained action code (32 states).

category

Character. Mid-level category (17 states).

superclass

Character. High-level superclass (6 states).

An object of class data.frame with 19347 rows and 9 columns.

An object of class data.frame with 19347 rows and 9 columns.

An object of class data.frame with 19347 rows and 9 columns.

An object of class data.frame with 10796 rows and 9 columns.

An object of class data.frame with 10796 rows and 9 columns.

An object of class data.frame with 10796 rows and 9 columns.

An object of class data.frame with 8551 rows and 9 columns.

An object of class data.frame with 8551 rows and 9 columns.

An object of class data.frame with 8551 rows and 9 columns.

An object of class data.frame with 429 rows and 164 columns.

An object of class data.frame with 428 rows and 138 columns.

Source

Saqr, M. (2026). Human-AI vibe coding interaction study.

Details

Nine long-format datasets are provided, filtered by actor and named by granularity level:

Dataset	Actor	Granularity
human_ai	Both	code (32 states)
human_ai_cat	Both	category (17 states)
human_ai_super	Both	superclass (6 states)
human_detailed	Human	code (32 states)
human_cat	Human	category (17 states)
human_super	Human	superclass (6 states)
ai_detailed	AI	code (32 states)
ai_cat	AI	category (17 states)
ai_super	AI	superclass (6 states)

Two wide-format datasets at category level (rows = sessions, columns = T1, T2, ...):

human_wide	Human actions in wide sequence format
ai_wide	AI actions in wide sequence format

Examples

# \donttest{
# Build a transition network from human category sequences
net <- build_network(human_wide, method = "relative")

# Use the edge list directly
head(human_ai_edges)
#>            from            to weight   session_id session   project order
#> 1       Context        Direct      1 0086cabebd15       1 Project_7     1
#> 2        Direct Specification      1 0086cabebd15       1 Project_7     2
#> 3 Specification     Interrupt      1 0086cabebd15       1 Project_7     3
#> 4     Interrupt      Delegate      1 0086cabebd15       1 Project_7     4
#> 5      Delegate          Plan      1 0086cabebd15       1 Project_7     5
#> 6          Plan  Verification      1 0086cabebd15       1 Project_7     6
#>                  timepoint from_actor to_actor from_category to_category
#> 1 2026-03-05T11:32:52.057Z      Human    Human       Specify     Command
#> 2 2026-03-05T11:32:52.057Z      Human    Human       Command     Specify
#> 3 2026-03-05T11:32:52.057Z      Human    Human       Specify   Interrupt
#> 4 2026-03-05T11:32:52.068Z      Human       AI     Interrupt    Delegate
#> 5 2026-03-05T11:32:56.945Z         AI       AI      Delegate        Plan
#> 6 2026-03-05T11:32:56.945Z         AI    Human          Plan      Verify
#>   from_superclass to_superclass
#> 1       Directive     Directive
#> 2       Directive     Directive
#> 3       Directive Metacognitive
#> 4   Metacognitive        Action
#> 5          Action        Repair
#> 6          Repair    Evaluative
# }