Kukui Bot · Engineering

How Kukui Bot builds software

A size-gated multi-agent pipeline: multi-model planning diversity, a single project manager who synthesizes the plan, a right-sized coder, and evidence-based QA. The default for real product work — not a rigid ritual for every typo.

PM · Grok 4.5 Plan · multi-model fan-out Dev · Grok by default QA · evidence, not trust Updated July 2026

01 Verdict

✓ Best default — medium+

Multi-model plan diversity → one owner synthesizes → fast default coder → independent evidence QA. Matches how a multi-provider agent platform should route work.

✗ Worst if rigid always

Full five steps for every typo burns latency and noise. Small work short-circuits. Huge work upgrades the coder — it does not force “always Grok + always four planners.”

Bottom line: The five-step diagram is the correct standard medium+ path. The best process is that path plus size gates, the right coder, and evidence-based PM QA.

02 Standard pipeline (medium+)

What a normal feature or fix looks like end to end.

1
Project manager

Clarify

Scope, options, brief. Brainstorm before a big dispatch.

2
Planning team

Fan-out

One prompt → several strong models in parallel (often Grok, Opus, Codex, Composer).

3
Project manager

Synthesize

Merge into one executable plan. Tradeoffs go to the human for architecture calls.

4
Developer

Implement

Default Grok. Auto-route to the best free developer. Keep phases short.

5
Project manager

Final QA

Re-run tests or inspect files. An evidence gate — not a rubber stamp.

What each step is for

StepJobWhy it exists
PM clarifies Owns scope, options, and the brief before anyone codes Prevents thrash
Planning fan-out Several models answer the same hard question in parallel Beats single-model blind spots
PM synthesizes One owner turns diversity into a single plan Avoids committee coding
Developer implements One coder ships against the plan Speed with ownership
PM final QA Independent check with fresh evidence “Done” is proven, not claimed

Common misconceptions

03 Size ladder (the real “best process”)

Best process = the pipeline plus these gates. Short-circuit small work; upgrade the coder for hard work.

Small Clear · ≤1 file · low risk · typo / one-liner / CSS
PM optional Dev (Grok / Composer / tiny) PM verify commit

Skip planning entirely. Prefer the cheapest capable model.

Normal Default feature / fix · multi-file OK · some ambiguity
PM clarify Planning fan-out PM synth plan Grok Dev PM QA + tests commit · brief

This is the standard product path.

Large Architecture · multi-phase · deep audit · high blast radius
PM clarify + options Planning fan-out PM phased plan Opus (or Grok) by phase PM gate each phase Final QA commit

Aim for short phases. Optional separate test phase. Humans decide architecture tradeoffs.

Stuck Stubborn multi-bug · deep root cause · prior attempts failed
Same plan path Codex implement / debug PM QA (careful severity check)

Depth is worth the wait. Not the default coder.

// One-liner PM owns truthmodels diversify the planone coder shipsPM proves it

04 Code-writing model order

When no model is forced, the platform prefers this order among free developer seats — never invents a missing model.

1
Grok 4.5
Default code — normal features & fixes
default
2
Claude Opus
Large / multi-file / long-context / architecture
heavy lift
3
Codex
Hardest debug only — stubborn multi-bug root cause
depth
4
Composer
Cheap standard / single-file work
efficient
5
Lightweight models
Tiny fixes only
micro
Routing rule: Prefer the best free open developer in order. If none of the primary coders are available, stop and open one — do not silently fall back to a weak seat.

05 Principles that make it work

1 Plan diversity, single owner

Multiple models plan in parallel; one PM synthesizes. Avoids committee coding and single-model blind spots.

2 Size gates

Full pipeline only when ambiguity or blast radius justifies it. Small work short-circuits.

3 Evidence QA

PM re-runs tests or inspects files. Developer “done” is a claim, not a fact.

4 Phase discipline

Keep dispatches short. Split implement vs test when both are non-trivial.

5 PM never ships production code

Coordinates, synthesizes, gates. Architecture options come with tradeoffs — the human decides.

6 Restarts are owned

Delegates do not bounce the server. Prefer hot-reload; full restart only when core files change.

Quality gate (before “done”)

Deliverables match the brief
Tests or verification evidence present (fresh, not remembered)
No silent scope expansion
Plain-language status for the operator — not a wall of code
Related code considered; neighbors not broken

06 Parallel tracks (not this chain)

Topic research
Dedicated multi-lens research pipeline — not the product planning team.
Markets
Public ticker and private-company deep dives run as their own pipelines.
Security audit
Broad scan first, deep verify second. Treat severity inflation carefully.
Infrastructure
IT Admin path — not the product PM → planning → developer chain.
Chat UX
Harness → integrity suite → feature branch → human approval before main.

07 Recommended default

Memorize this

SMALL PM? → Dev → verify → commit NORMAL PM clarifyPlanning×NPM planGrok DevPM QA → commit LARGE PM + options → Planning×N → phased plan → Opus/Grok by phase → gates → final QA STUCK same plan path → Codex → careful QA

Run this process on your machine

Kukui Bot is self-hosted multi-model agents with real tools, delegation, and evidence-based workflows.