/process
How we
actually work.
The full delivery process — the thesis behind agent-first development, the shape of the 720-hour cycle, where humans intervene, how we handle security in regulated verticals, and the engagements we deliberately decline.
a note from the founder
This is a note from the founder.
I started kaedax because I had spent too many years watching software shops quote six months for work I knew could ship in thirty days. The pivot wasn't a clever insight. It was accepting that the typing bottleneck had moved, and building a studio around that fact instead of pretending it hadn't.
Every engagement starts with a founder on the call. We sign the spec. We review the ADRs. Our number is in the runbook you walk away with. When founders ask "who actually shows up?" the answer is a founder of kaedax, and the engineer who'll lead your build, sitting next to them.
We're opinionated builders. If we work together, you'll get strong points of view from us on the things we've seen go wrong before. You'll also get a team that knows when to shut up and ship.
Between the founders, we bring more than fifty years of cumulative engineering experience. That spans high-scale consumer products, regulated platforms, and the kind of internal tooling that gets written off as boring until it doesn't work. We've seen a lot of patterns fail. We've also watched the new ones quietly become the default. Kaedax is the bet that the next ten years belong to teams built around the second list.
If any of that lands, send a brief. The address is hello@kaedax.com, our inbox, and a founder reads every one.
— A founder of kaedax
kaedax · electronic city, bengaluru
[ §01 ] the thesis
Engineers stopped
being the bottleneck.
For thirty years the software-services industry was sized to a single constraint: how many engineers could you put in a room, and how quickly could they type. Every process artefact — the project plan, the weekly status, the four-month timeline — was a workaround for that constraint.
The constraint has moved.
Engineers are no longer the typing bottleneck. They are now the *judgement* bottleneck. What kaedax does is take that observation seriously: agents own the work where typing is the bottleneck, and humans own the work where judgement is. The result isn't a faster agency. It's a differently-shaped one.
A 720-hour cycle is what naturally falls out. Long enough to ship a real product; short enough that compounding judgement calls don't drift; honest enough that we can quote it without padding.
[ §01.5 ] how it starts
From brief
to cycle.
Every kaedax engagement starts the same way. No discovery theatre, no proposal padding — two weeks from your first message to the start of the cycle.
▷ total: ~14 days · brief → cycle start
- 01
Brief, sent over
Hour 00 → Hour 24You write us — even one paragraph is fine. We respond within 24 hours from a real engineer, with a yes / not-now / not-us. No sales sequences, no forms.
↳Reply lands in your inbox · signed by a founder
- 02
Scope call, 45 minutes
Week 01A founder of kaedax and the engineer who'd lead your build. You walk us through the product; we tell you on the same call whether it fits a 720-hour cycle. We say no to roughly a third of inquiries here.
↳Outcome: a one-page fit memo · sent within 48h
- 03
Discovery sprint, 5 working days
Week 02Mutual NDA, then a tight discovery: spec.md draft, ADR shortlist, architecture sketch, risk register. At the end of the sprint we share a fixed-scope, fixed-timeline engagement proposal — or we recommend you don't proceed.
↳Artifact: spec.md v0 · yours regardless of next step
- 04
720 hours, on the clock
Cycle startsKickoff Monday. Agents and humans into the same repo. Daily cycle log streaming to your team. Two senior human engineers reviewing every PR. Cutover at hour 720.
↳Followed by 30 to 60 days of post-launch on-call
24h
Response from a founder, not a form
5d
Discovery sprint produces spec.md you keep
≤ 30%
Of inquiries we say yes to
[ §02 ] the loop
Six agents,
four phases.
A simple matrix — each phase, who owns it, who co-owns it. Every agent has its own eval set, its own runbook, and its own clear boundary. Below is what runs in every single engagement.
| Phase | scope | build | qa | deploy | monitor | ops |
|---|---|---|---|---|---|---|
| Scope | owns | — | — | — | — | drafts |
| Build | watches | owns | co-owns | — | — | drafts |
| Polish | — | co-owns | owns | preps | wires | drafts |
| Cutover | — | supports | — | owns | owns | drafts |
● "owns" = primary responsibility · "co-owns" = paired with human reviewer · em-dash = idle in this phase
[ §03 ] human moments
Exactly where
humans step in.
Agent-first does not mean human-absent. The points below are the non-negotiable human-in-the-loop checkpoints. Everything else is the agents earning their keep.
- Day 1 Founder kickoff kaedax founder + your founder(s) + lead engineer
- Day 3 Spec signoff Founders ratify spec.md; ADRs approved by your CTO
- Every PR Code review Two senior human engineers approve before merge
- Day 14 Mid-cycle review Architecture review with your team; scope drift named
- Day 22 Polish gate Founder taste-test of every surface; copy + flows
- Day 27 Cutover signoff Joint go/no-go between kaedax + your team
- Post-launch 30-60d on-call kaedax engineer responds to incidents in-hours
[ §04 ] security · IP · data
Your code, your cloud,
your secrets.
[01]
Mutual NDA at zero cost
Signed before scope discussions. Default term: 24 months, perpetual on trade secrets.
[02]
Repo in your accounts
GitHub / GitLab in your org. We get scoped seats, removable in one click. No code on kaedax laptops at end of cycle.
[03]
Secrets stay in your vault
AWS Secrets Manager, Doppler, 1Password — your choice. kaedax engineers see what they need to ship, and only that.
[04]
Agents in your CI
Build, QA, deploy agents run in your CI/CD against your vaults. Logs go to your observability stack — we have read-only access during the cycle.
[05]
Eval harness on AI surfaces
Any AI feature we ship comes with an evaluation harness — golden examples, thresholds, regression alerts. You own the harness.
[06]
Customer-managed encryption
For regulated verticals (PHI, PII, payments), CMK by default. We design for it; the keys belong to you.
[ §05 ] verticals
Where we go
deep.
We don't claim domain expertise we don't have. Below are the verticals we've shipped into and the specific posture we bring to each.
What we ship · Underwriting consoles, embedded payments, treasury workflows, audit-grade decisioning
Posture · PCI-aware controls, append-only audit trails, decision explainability UIs, regulator-ready evidence trails
What we ship · Clinician consoles, telehealth surfaces, AI scribes with eval harnesses, patient workflows
Posture · PHI handling defaults, ephemeral inference, CMK encryption, BAA-ready architecture, eval corpora as first-class artefacts
What we ship · Embedded insurance APIs, partner consoles, policy-lifecycle workflows, claims-adjacent (not claims)
Posture · Contract-first APIs, regulator-aware SLAs, signed audit trails, IRDAI/state-DOI evidence-readiness
What we ship · DTC storefronts, configurators with real depth, headless commerce, subscription products
Posture · PCI-respectful checkout, GA4 + server-side events, performance budgets, GDPR/DPDP-respectful defaults
What we ship · Community products, creator platforms, social surfaces, habit/behaviour apps
Posture · Moderation defaults from day one, social-spec discipline, trust + safety baked in, live-ops agents at launch
[ §06 ] where we work from
Bengaluru,
building globally.
Kaedax is headquartered in Electronic City, Bengaluru. Our cycles overlap with North America, Europe, and Asia-Pacific time zones; founders we work with in London, San Francisco, Singapore, and Mumbai all meet kaedax engineers in their own working hours.
We are a studio, not an offshore vendor. The same people who scope your engagement are the ones who review the PRs, sign the ADRs, and answer the phone post-launch.
[ §07 ] what we won't do
The engagements
we decline.
Being explicit about this is more useful than any marketing claim. We say no to roughly a third of inquiries — usually on the scope call.
-
[01]
Multi-product platform builds. The 720-hour cycle does not extend.
-
[02]
Pure design or pure strategy engagements. We ship code; design and strategy emerge from the spec process, not as standalone deliverables.
-
[03]
Regulated builds where the founder wants kaedax to be the auditor of record. We're the engineering partner; we work alongside your compliance counsel — not in place of them.
-
[04]
Engagements where the founder wants to outsource taste. Agent-first development assumes a strong human point of view on what the product is.
-
[05]
White-label work that obscures who built what. We won't ship under another agency's name.
[ §08 ] operating principles
Concrete pledges.
Hold us to them.
Six commitments that govern how kaedax operates. Each is testable. Each has a consequence if we break it. We update this list when our practice changes; we do not edit it to flatter ourselves.
- [01]
24-hour response from a founder
Every inbound brief gets a reply within 24 working hours, written by a founder, with one of three answers: yes / not-now / not-us. No sales sequences. No forms longer than four fields.
- [02]
No hourly billing, ever
Every engagement is fixed-scope, fixed-timeline. If we estimate wrong, we eat the overrun. Hourly billing rewards the shop for being slow; we are not interested in that economics.
- [03]
Your code, your cloud, your secrets
Repo lives in your accounts. Secrets in your vault. Deploy keys you control. We leave with no continuing access to anything except an off-boarding email.
- [04]
We sign the ADRs we approve
Architecture decision records are signed by name — by the kaedax engineer who proposed them and the founder who approved them. Pseudonyms or 'the team' do not appear on documents that will outlive the engagement.
- [05]
We publish what we won't take
Engagements we decline are listed publicly on this page. Founders know up front whether kaedax is the right fit before they spend an hour on a call.
- [06]
Runbooks that don't require us
We leave the engagement with operations runbooks written for an engineer who has never met us. If you can't run the system without kaedax on retainer, we have not done our job.