One-Shot Shop Challenge: Building a Full Shop with AI Coding Agents

All Builds at a Glance

Newest first. Click a column header to sort, click a build to jump to its details. The feature score is weighted: PASS = 1, PARTIAL = 0.5, out of 143 tests.

Build	Date▾	Mode	Feature Score▾	Cost▾	Time▾
#15Codex GPT-5.6 Sol Ultra	Jul 12	Team	95.8%	$120	2h 55m
#14Claude Code Fable 5	Jun 9	Sub-agents	99.7%	$423	6h 58m
#13Claude Code Team 4.8 xHigh	May 30	Team	95.5%	$498	2h 54m
#12Codex GPT-5.5 Goal Mode	May 4	Sub-agents	92.0%	$530	23h 50m
#11Codex GPT-5.5	Apr 25	Sub-agents	82.9%	$19	48m
#10Claude Code Opus 4.7 xHigh	Apr 16	Team	43.0%	$157	1h 36m
#09Claude Code Opus 4.7	Apr 16	Single	73.4%	$23	32m
#07Claude Code Team v4	Mar 20	Team	89.9%	$132	3h 39m
#06Claude Code Team v3	Mar 18	Team	87.8%	$285	10h 59m
#05Codex with Sub-Agents v2	Feb 14	Sub-agents	58.0%	$28	3h 27m
#04Claude Code Team v2	Feb 13	Sub-agents	66.8%	$74	3h 00m
#02Claude Code with Sub-Agents	Feb 13	Sub-agents	60.5%	$62	2h 13m
#03Codex with Sub-Agents	Feb 12	Sub-agents	65.7%	$9	1h 44m
#01Claude Code with Team Mode	Feb 11	Team	90.6%	$73	1h 06m

The Experiment

Same spec, same tech, same single prompt. Only the coding agent changes. Every run is scored against the same 143 end-to-end acceptance tests, and everything is published: prompts, session logs, analysis, and code.

The Setup

Specification

Full spec with acceptance criteria, no code snippets

tecsteps/shop/.../specs

Codebase

Fresh Laravel template with Livewire

MCP Servers

Laravel Boost, Playwright

QA Verification

143 E2E tests, two independent checks per build

View Testplan

The Builds

Newest first. Each build links to its full story, session analysis, code quality report, feature results, and database schema.

#15 Codex GPT-5.6 Sol Ultra(Codex CLI v0.144.3, gpt-5.6-sol-pro, multi-agent v2 team mode)

Codex CLI v0.144.3gpt-5.6-sol-proMulti-Agent v216 Named Sub-AgentsTeam Mode

16 named sub-agents role-playing a full engineering org. Highest feature pass rate of any Codex run, and remarkably cheap for its completeness: 98% of input was cache reads, so 16 agents cost only ~$120. Repeats the cross-tenant admin data leak. Read the full build story →

2h 55m

Duration

$119.74

API Cost

286

Classes

Active Agents

Feature Tests: 135 of 143 passed (95.8%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$119.74

Time

2h 55m

Feature Completeness

out of 143 tests

Pass

135

Partial

Fail

Code Quality

LOC

11,292

Code smells

139

Tech debt

26.9 h

Duplication

0.8%

Teammates

teammates

Count

#14 Claude Code Fable 5(Claude Code v2.1.170, Fable 5 with high reasoning, orchestration left to the agent)

Claude Code v2.1.170Fable 5High ReasoningAgent's ChoiceSub-Agents

Fable 5 free to choose its own orchestration self-organised into a 14-phase sub-agent pipeline. Highest feature score on the page (99.7% weighted) and a polished, feature-rich result, at the cost of the longest active runtime here. Read the full build story →

6h 58m

Duration

$422.68

API Cost

195

Classes

Active Agents

Feature Tests: 142 of 143 passed (99.7%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$422.68

Time

6h 58m

Feature Completeness

out of 143 tests

Pass

142

Partial

Fail

Code Quality

LOC

10,314

Code smells

Tech debt

12.6 h

Duplication

1.5%

Teammates

sub-agents

Count

#13 Claude Code Team 4.8 xHigh(Claude Code v2.1.158, Opus 4.8 with xHigh reasoning, thinking on, 1M context, strict team mode)

Claude Code v2.1.158Opus 4.8xHigh ReasoningThinking On1M ContextTeam Mode

Strict team-mode run: a 6-teammate team building the whole shop from a single "do it in one go" brief. Best maintainability on the page, with only address-book and stock-cap gaps left; cost sits in the upper range. Read the full build story →

2h 54m

Duration

$497.67

API Cost

205

Classes

Active Agents

Feature Tests: 134 of 143 passed (95.5%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$497.67

Time

2h 54m

Feature Completeness

out of 143 tests

Pass

134

Partial

Fail

Code Quality

LOC

11,230

Code smells

Tech debt

10.5 h

Duplication

0.8%

Teammates

teammates

Count

#12 Codex GPT-5.5 Goal Mode(OpenAI Codex CLI v0.128.0, GPT-5.5 with xHigh reasoning, persistent goal mode)

Codex CLI v0.128.0gpt-5.5xHigh ReasoningPersistent GoalPlan ModeSub-AgentsCodex Pro

23 hours of autonomous work driven by a single goal brief. Strong pass rate, but also the largest codebase and the longest tech-debt tail. Read the full build story →

23h 49m

Duration

$530.26

API Cost

268

Classes

Active Agents

Feature Tests: 126 of 143 passed (92.0%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$530.26

Time

23h 50m

Feature Completeness

out of 143 tests

Pass

126

Partial

Fail

Code Quality

LOC

13,645

Code smells

Tech debt

23.2 h

Duplication

3.2%

Teammates

sub-agents

Count

#11 Codex GPT-5.5(OpenAI Codex CLI v0.124, GPT-5.5 with high reasoning)

Codex CLI v0.124.0gpt-5.5High ReasoningSub-AgentsCodex Pro

Smallest codebase and lowest cost on the page; clean SonarCloud gate but skipped many admin features. Read the full build story →

47m 38s

Duration

$18.85

API Cost

Classes

Active Agents

Feature Tests: 108 of 143 passed (82.9%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$18.85

Time

48m

Feature Completeness

out of 143 tests

Pass

108

Partial

Fail

Code Quality

LOC

1,743

Code smells

Tech debt

2.0 h

Duplication

1.3%

Teammates

sub-agents

Count

#10 Claude Code Opus 4.7 xHigh(Same setup as #09, stricter prompt to enforce team-mode)

Claude Code v2.1.114Opus 4.7xHigh ReasoningThinking On1M ContextTeam Mode

Same setup as #09 with a hardened prompt that forced team-mode; the team ran and over-built the spec on paper, but buttons shipped dead. Read the full build story →

1h 36m

Duration

$157.31

API Cost

151

Classes

Active Agents

Feature Tests: 51 of 143 passed (43.0%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$157.31

Time

1h 36m

Feature Completeness

out of 143 tests

Pass

Partial

Fail

Code Quality

LOC

5,043

Code smells

Tech debt

3.6 h

Duplication

1.4%

Teammates

teammates

Count

#09 Claude Code Opus 4.7(Same prompt as #01, latest Opus)

Claude Code v2.1.112Opus 4.7High ReasoningThinking On1M Context

Opus 4.7 ignored team-mode and built everything as a single agent - the fastest and cheapest Claude run. Read the full build story →

32m 27s

Duration

$22.63

API Cost

105

Classes

Active Agents

Feature Tests: 93 of 143 passed (73.4%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$22.63

Time

32m

Feature Completeness

out of 143 tests

Pass

Partial

Fail

Code Quality

LOC

2,667

Code smells

Tech debt

2.6 h

Duplication

0.4%

Teammates

single

Count

#07 Claude Code Team v4(Same Prompt, 1M Context)

Claude Code v2.1.81Team ModeOpus 4.6High Reasoning1M Context

Same prompt as #01 on 1M context; stable specialists and the highest feature pass rate of the Opus 4.6 builds. Read the full build story →

3h 39m

Duration

$132.06

API Cost

389

Files Created

Active Agents

Feature Tests: 121 of 143 passed (89.9%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$132.06

Time

3h 39m

Feature Completeness

out of 143 tests

Pass

121

Partial

Fail

Code Quality

LOC

4,537

Code smells

Tech debt

8.4 h

Duplication

1.4%

Teammates

teammates

Count

#06 Claude Code Team v3(Advanced Prompt, 1M Context)

Claude Code v2.1.80Team ModeOpus 4.6High Reasoning1M Context

Advanced prompt with a controller and QA teammate - broadest coverage of the early builds but the longest and most expensive run of its generation. Read the full build story →

10h 59m

Duration

$284.52

API Cost

482

Files Created

158

Active Agents

Feature Tests: 119 of 143 passed (87.8%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$284.52

Time

10h 59m

Feature Completeness

out of 143 tests

Pass

119

Partial

Fail

Code Quality

LOC

5,708

Code smells

Tech debt

14.3 h

Duplication

8.0%

Teammates

teammates

Count

159

#05 Codex with Sub-Agents v2(More Instructions)

OpenAI Codex v0.101.0Sub-Agents (experimental)GPT-5.3-codexReasoning: xhigh

Codex rerun with a quality-focused prompt and xhigh reasoning; lots of classes, lots of tech debt. Read the full build story →

3h 27m

Duration

$28.40

API Cost

Agents Spawned

898

Tool Calls

Feature Tests: 70 of 143 passed (58.0%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$28.40

Time

3h 27m

Feature Completeness

out of 143 tests

Pass

Partial

Fail

Code Quality

LOC

7,178

Code smells

113

Tech debt

25.1 h

Duplication

3.0%

Teammates

sub-agents

Count

#04 Claude Code Team v2(More Instructions)

Claude Code v2.1.41Sub-AgentsOpus 4.6Thinking: OnReasoning: Max

Tuned prompt with explicit review agents; cleanest code-smell profile of the sub-agent runs. Read the full build story →

3h 0m

Duration

$73.92

API Cost

376

Files Created

Active Agents

Feature Tests: 82 of 143 passed (66.8%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$73.92

Time

3h 0m

Feature Completeness

out of 143 tests

Pass

Partial

Fail

Code Quality

LOC

6,033

Code smells

Tech debt

5.2 h

Duplication

3.8%

Teammates

sub-agents

Count

#02 Claude Code with Sub-Agents

Claude Code v2.1.41Sub-AgentsOpus 4.6Thinking: On

Same prompt as #01 but with sub-agents instead of teammates; slower and less complete. Read the full build story →

2h 13m

Duration

$61.97

API Cost

358

Files Created

Active Agents

Feature Tests: 73 of 143 passed (60.5%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$61.97

Time

2h 13m

Feature Completeness

out of 143 tests

Pass

Partial

Fail

Code Quality

LOC

6,033

Code smells

Tech debt

8.6 h

Duplication

3.6%

Teammates

sub-agents

Count

#03 Codex with Sub-Agents

OpenAI Codex v0.99.0Sub-Agents (experimental)GPT-5.3-codexReasoning: xhigh

First Codex pass at the same challenge - by far the cheapest run, but missing many features. Read the full build story →

1h 44m

Duration

$8.79

API Cost

Agents Spawned

357

Tool Calls

Feature Tests: 89 of 143 passed (65.7%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$8.79

Time

1h 44m

Feature Completeness

out of 143 tests

Pass

Partial

Fail

Code Quality

LOC

6,037

Code smells

Tech debt

12.7 h

Duplication

2.8%

Teammates

sub-agents

Count

#01 Claude Code with Team Mode

Claude Code v2.1.39Team ModeOpus 4.6Thinking: On

Baseline run: Opus 4.6 in team mode, the reference point the other builds are compared against. Read the full build story →

1h 6m

Duration

$73.44

API Cost

388

Files Created

Active Agents

Feature Tests: 126 of 143 passed (90.6%)

Live Shop Admin Panel Source Code

Build Detail Quality Features Schema

Admin login:

Efficiency

shorter = better

Cost

$73.44

Time

1h 6m

Feature Completeness

out of 143 tests

Pass

126

Partial

Fail

Code Quality

LOC

6,108

Code smells

168

Tech debt

22.3 h

Duplication

2.9%

Teammates

teammates

Count