01
● online scope.agent
From brief to spec
Ingests your founder brain-dump, runs it through clarifying questions, drafts a working spec, and flags ambiguity before it becomes rework.
spec.md ADR drafting open-question log
Q3 intake · 1 of 2 slots open
/ est. 2026 · global delivery
v0.1 · the kaedax index
an ai-native software agency
Agent-first software, shipped in
Most agencies use AI to type faster. We don't. Our agents run the delivery loop end-to-end, which is why thirty days is a real deadline and not an aspirational one. We work out of Bengaluru with founders who cannot wait six months.
720h
Cycle length
2 / cycle
Engagements accepted
100%
Code & agents yours
▷ live cycle log build #027 · day 14 / 30
cycle · livekaedax-cycle · live agents
● online streamed from a real kaedax build
◆scoped, signed, shipped — under NDA ◆you own the repo from day one ◆agents do the toil — humans do the judgement ◆thirty days · 720 hours · 43,200 minutes ◆ship · monitor · iterate · repeat ◆two engagements per cycle · curated intake ◆bengaluru hq · operating globally ◆scoped, signed, shipped — under NDA ◆you own the repo from day one ◆agents do the toil — humans do the judgement ◆thirty days · 720 hours · 43,200 minutes ◆ship · monitor · iterate · repeat ◆two engagements per cycle · curated intake ◆bengaluru hq · operating globally
[ §00.5 ] what you can hold us to
A working MVP in 30 days.
Fixed from $15K.
Not an estimate, not a range that drifts. A real, deployable product — in your cloud, owned by you — quoted at a fixed number before we write a line of code. The agents carry the volume; senior humans hold the bar.
- 15+
-
products shipped
real, deployed software across regulated and consumer verticals
- 30 days
-
to a working MVP
720h, agent-accelerated — guaranteed delivery, not a prototype
- from $15K
-
fixed price
quoted before kickoff · no hourly meter, no scope-creep invoices
- 100%
-
yours
code, repo, cloud and keys in your accounts · zero lock-in
- 2 / cycle
-
engagements at a time
curated intake · founders only · senior humans on every PR
[01]
[02]
[03]
[04]
[05]
▷ posture
- —Not a dev shop.
- —Not an offshore vendor.
- —Not an account-managed agency.
A studio where agents do the typing and humans hold the bar. The first call is with the founder and the engineer who'd lead the build. No proposals from people who won't show up later. No handoffs to a delivery team in another room.
- ● Founding team · 50+ years cumulative engineering
- ● Electronic City, Bengaluru · working across SF · NYC · London · Singapore
[ §01 ] the problem
Your agency quoted six months.
You don't have six months.
Bootstrapped founders aren't short on ideas. They're short on runway. The traditional agency model (scope phases, weekly check-ins, slow handovers) was built for a world where engineers were the bottleneck. They aren't anymore, and we think the agencies that adapt last will lose this decade.
kaedax is what happens when you accept that the agents are the engineers now, and the humans are here to make taste calls.
01
6 mo
What the agency quoted
02
$180k
What the agency wants
03
11 ppl
Slack channels you'll be in
04
47
Status meetings you'll attend
time-to-ship · same scope
agency
~ 4,320 hours
kaedax
6× faster · same depth
[ §02 ] the 720h cycle
What thirty days
actually looks like.
No mystery, no Gantt theatre. Four phases, four artifacts, one fixed deadline.
-
Day 01 — 03
T+000 → T+072
Brief → spec
You talk to a human. We talk to scope.agent. By day 3 there's a signed spec, an ADR log, and a Figma flow — not a 40-slide proposal.
↳spec.md ↳adr/ ↳ux/flows.fig -
Day 04 — 21
T+073 → T+504
Agent-driven build
Build, QA, deploy agents run on a tight loop. Two human engineers steer, review every PR, and own the hard judgement calls. You watch the cycle log in real time.
↳main · 87 PRs ↳vitest · 91% cov ↳preview deploys × 14 -
Day 22 — 27
T+505 → T+648
Polish + integration
The unglamorous week. Edge cases, error states, onboarding copy, perf budgets, the analytics you forgot to ask for. This is the part shops skip.
↳lighthouse 99+ ↳sentry armed ↳stripe live -
Day 28 — 30
T+649 → T+720
Handoff + ship
Production cutover. You get the repo, the agents, the runbooks, the dashboards. We're on call for 30 days after — no extra invoice.
↳prod live ↳runbook.md ↳30d on-call
[ §03 ] the agent stack
Six agents.
One delivery loop.
Each agent owns one part of the cycle. None of them try to be the whole team. The humans do what humans are still better at: taste, judgement, and the awkward call.
02
● online build.agent
Module-level engineering
Owns modules end-to-end — schema, API, UI, tests, docs. Opens PRs against a human reviewer. Won't merge to main without a green QA pass.
repo writes PR drafts self-review
03
● online qa.agent
Tests, perf, regression
Generates tests from the spec, not from the code. Runs the suite on every PR. Tracks coverage, lighthouse, and a perf budget you'll actually feel.
vitest gen playwright runs perf budget
04
● online deploy.agent
Preview → prod cutover
Provisions preview environments on every PR, handles env vars, runs migrations safely, and executes the production cutover with a rollback plan.
vercel / fly migrations rollback plan
05
● online monitor.agent
Sentry · Axiom · uptime
Wires telemetry from day one. Triages incoming errors, deduplicates them, opens issues in the repo, and pages a human only when it matters.
sentry axiom cron pings
06
● online ops.agent
Standups, status, handoff
Writes the daily standup, the weekly status, the runbook, and the handoff doc — so the team spends zero hours on agency theatre.
daily digest runbook weekly note
Agent — owns a phase
Artifact — produced or consumed
Live cycle stream
[ §03 · live ] the guardrail is the product
Agents that run your product — and stop when they should.
Not a chatbot bolted on the side. The agents we ship operate inside the product, call real tools, and pause for a human on anything that moves money, data, or infrastructure. When policy says no, they refuse — and say why.
- Streamed plan + MCP tool calls — you see what it's about to do, and what it touched
- Human-in-the-loop approval — no irreversible action runs without a sign-off
- Eval-gated, policy-aware — the same refusal layer guarding every vertical demo
→ try it in any live demo · look for the ✦ copilot
✦build.agent
· eval-gated · human-approved
liveRefund order #4471 and notify the customer
replay a run
[ §04 ] recent work
Selected
engagements.
Five recent cycles across fintech, healthtech, ecom, insurance, and B2C — anonymized under NDA. The engineering shape and outcomes are as delivered.
shapes we take: MVPs · domain platforms · agent products
-
AI · Consumer agent product 01 / 06 codename
TALLOW
An end-to-end personal-finance agent for solopreneurs — perception, reasoning, memory, tools, and an eval harness — shipped as a product, not a demo.
TALLOW set out to build a 'personal CFO' agent for solopreneurs: it ingests bank feeds, classifies transactions, surfaces decisions, and answers natural-language questions about ru…
Lisbon · PT · 1 cycle · 720hopen → -
Fintech · SMB credit 02 / 06 codename
LATTICE
An underwriting console for an SMB credit fintech, built and shipped in a single cycle.
LATTICE needed to replace a spreadsheet-driven credit committee with an internal underwriting console that pulled bureau data, ran their proprietary scoring model, and let analysts…
Singapore · APAC · 1 cycle · 720hopen → -
AI · Embedded feature 03 / 06 codename
AURORA
An AI agent module dropped into an existing B2B SaaS, behind a feature flag, with an eval harness that gated every PR.
A workflow SaaS with 1,200 enterprise customers wanted to ship an AI assistant inside their product without disrupting the existing surface or risking regression. We built the agen…
Toronto · CA · 1 cycle · 720hopen → -
Healthtech · Chronic care 04 / 06 codename
CADUCEUS
A clinician-facing chronic-care platform with patient intake, structured encounter notes, and an AI scribe that earns clinician trust.
CADUCEUS needed to ship a working clinician console — intake, scheduling, encounter notes, an AI-assisted scribe, and an audit trail — for a pilot with two clinics. Two cycles, seq…
Bay Area · USA · 2 cycles · 1,440hopen → -
B2C · Creator economy 05 / 06 codename
PULSE
A community-first creator platform that turned a 4,000-person waitlist into a launch, with the agents quietly running the live ops layer.
PULSE is a creator-community product — a hybrid of a paid newsletter, a member-only feed, and a private chat. The founders had a 4,000-person waitlist and twelve weeks before a ref…
Berlin · DE · 1 cycle · 720hopen → -
Ecommerce · DTC home 06 / 06 codename
BOUGH
A direct-to-consumer storefront with a made-to-order configurator, replacing a Shopify build that had outgrown the platform.
BOUGH sells solid-oak furniture made-to-order. Their Shopify store buckled at the configurator step — too many options, too many SKUs, too much pricing logic. We rebuilt the storef…
London · UK · 1 cycle · 720hopen →
● identifiers anonymized · published with explicit consent
all engagements →[ §04.5 ] in their words
What founders say
after the cycle.
founders & operators · roles as delivered
We expected a prototype. We got a production console with audit trails our compliance team was actually happy with. Three days early.
Every agency we spoke to treated HIPAA as a change order. kaedax treated it as the spec. The eval harness on the scribe is the reason our clinicians trust it.
Thirty days from a brief to a configurator our old agency said needed two quarters. Conversion went up 38% in the first month.
[ §05 ] how we work
A studio,
not a shop.
Headquartered in Bengaluru. Curated intake — two engagements per cycle, founders only. Built for ambitious teams in regulated and unregulated verticals alike. Below is what every engagement looks like before code is written.
Mutual NDA before scope
Nothing about your product, your team, or your conversations leaves the room without your written consent. We have turned away engagements rather than weaken this default.
Code lives in your cloud
Repo, CI, secrets, deploy keys: all in your accounts, never ours. We are guests on your infrastructure for the duration of the cycle, and we leave it cleaner than we found it.
Regulated-vertical posture
Scoped IAM, customer-managed keys, signed audit trails, an eval harness for any AI outputs. Table-stakes for fintech, healthtech, and insurance engagements, and where we expect to spend extra cycle time.
Human on every judgement call
Two senior engineers review every PR. Founders sign the ADRs. The agents do the volume; humans hold the bar, and we decide together when the bar should move.
▷ verticals · posture brought to each
- FINTECH PCI-aware controls · audit-grade trails →
- HEALTH PHI handling · ephemeral inference →
- INSURANCE Regulator-aware SLAs · evidence trails →
- ECOM PCI · DPDP · GDPR-respectful defaults →
- B2C Moderation defaults · trust + safety from day one →
+ ai/agents · marketplaces · internal tools · proptech · legaltech · education — all 11 verticals →
For the rest — the architecture, the human moments, the things we won't take on — read the full process.
read the full process →[ §06 ] objections
Things founders
ask first.
Honest answers. If your question isn't here, bring it to the scope call — we'd rather answer it directly.
01 Is thirty days actually enough?
+
Is thirty days actually enough?
For a focused MVP, a contained AI feature, or a domain-shaped product (one fintech surface, one healthtech workflow): yes. The agent loop is designed around that exact scope. For multi-product platforms it isn't, and we tell people so on the scope call. We do not take builds we can't ship.
02 What if the agents write bad code?
+
What if the agents write bad code?
Two senior human engineers review every PR. Nothing merges to main without a human approving the change and a green test suite. The agents do the volume; the humans hold the bar. You meet both on day one, before any code is written.
03 Do I own the code, the agents, and the prompts?
+
Do I own the code, the agents, and the prompts?
Yes. Entirely. Repo, agents, prompts, runbooks, eval harness, deploy keys. You leave the cycle with all of it, and we have no continuing access to anything once the post-launch on-call window ends. We do not gate anything behind a kaedax dependency.
04 How do you handle confidentiality and IP?
+
How do you handle confidentiality and IP?
Mutual NDA before any scope conversation. Code lives in your repo, in your cloud, under your accounts. Never ours. Secrets stay in your vaults. We publish case studies only with explicit written consent, and we anonymize client identifiers by default.
05 Can you work on regulated products in health, finance, or insurance?
+
Can you work on regulated products in health, finance, or insurance?
Yes. Most of our work sits in those verticals. We bring the security posture (encrypted-at-rest, scoped IAM, signed audit trails, an eval harness for any AI outputs) and we partner with your compliance counsel on the certifications. We are an engineering partner, not an auditor of record.
06 Which models and frameworks do you use?
+
Which models and frameworks do you use?
Defaults: Claude 4.6 or 4.7 as the planner model. MCP for tool contracts. LangGraph where we need explicit graph orchestration; otherwise narrow, direct code. pgvector for memory in most builds (we prefer the ops simplicity of Postgres). Inngest or Temporal for long-running jobs. We do not maintain a buffet of 'we work with everything.' We have opinions, we ship with them, and we deviate when there's a real engineering reason. Switching defaults mid-cycle is a deliberate ADR, not a slide.
07 What's the engagement size you work with?
+
What's the engagement size you work with?
Shared on the scope call once we understand what you are building. The cycle unit is fixed at 720 hours; the number of cycles depends on the product. We don't publish pricing because every engagement is shaped to its specific work, and because the founders we want to work with care less about a number than about who is actually building.
08 Who shows up on the call?
+
Who shows up on the call?
A founder of kaedax plus the engineer who would lead your build. Not a salesperson, not an account exec. If we can't help you we say so on the same call, and we usually point you at someone who can.
[ §07 ] field notes
What we learn,
written down.
engineering 01 / 03
MCP is the right contract for agent tools
Tool calling done as ad-hoc per-app function definitions is a fragility tax. Model Context Protocol is the contract layer the agent-product economy needed. Here is why we default to it, and the one case where we don't.
Jun 06, 2026
read →
manifesto 02 / 03
Why 720 hours is the right delivery window
Six months is a hedge, not an estimate. Two weeks is bravado. Thirty days is the only window that respects both founder runway and engineering reality.
Jun 02, 2026
read →
manifesto 03 / 03
The refusal layer is the product
Most agent demos fail at refusals, not at execution. For agents that touch money, health, or sensitive data, what the agent will not do is a sharper product surface than what it will. A note on building refusal-first.
May 30, 2026
read →