Doctrine & intent
How we think.
The doctrine behind the decisions we guide. 9 ideas, formed through delivery, architecture and engineering work, that shape how we approach the job — field-drawn, not immutable laws, but the synthesis of what we have shipped.
Foundational doctrine
The beliefs everything else hangs from.
- 01
Conventional wisdom: AI accelerates delivery by helping teams generate more output.
Once generation is abundant, acceptance is the constraint.
When AI makes generating code cheap, the scarce, valuable work becomes judging whether a change is correct, safe and worth shipping. Delivery should be optimised — and measured — around acceptance, not output.
Read the evidence → - 02
Conventional wisdom: Pick the best model and you win.
The model is rented; the harness is owned.
The durable competitive asset in agentic delivery is the harness — tests, evals, review patterns and a versioned context layer — that lets generated change be judged correct fast. The model is a commodity input.
Read the evidence → - 03
Conventional wisdom: Better process — Agile, SAFe, Scrum — improves delivery.
Alignment beats methodology.
Agile fixed cadence and left the seams between product, architecture and engineering un-owned. The constraint is the three-way alignment of intent, not the ceremony.
Read the evidence →
Positions that follow
Downstream consequences of the doctrine.
- 01
Conventional wisdom: AI saves money by cutting developer headcount.
Cheap generation creates expensive ownership.
When generation costs nothing, the bill moves downstream to acceptance, integration, review and ownership — and usually gets bigger. Budget the gap, not the headcount.
Read the evidence → - 02
Conventional wisdom: More output means more productivity.
Measure what lands, not what is typed.
Under abundant generation, volume metrics (lines of code, "percent faster") are misleading. The signals that matter are acceptance rate, rework and stability.
Read the evidence → - 03
Conventional wisdom: AI quality comes from better prompts and bigger context windows.
The context layer is the architecture.
Quality in agentic delivery comes from the persistent, version-controlled context layer you own — conventions, decisions and contracts — not the ephemeral prompt window your tool rebuilds each call.
Read the evidence → - 04
Conventional wisdom: Architecture is the diagrams and the technology stack.
Architecture is decision-making under uncertainty.
Diagrams and stack choices are the residue of architecture; the work is making and recording trade-offs under uncertainty, optimising for reversibility.
Read the evidence → - 05
Conventional wisdom: If the agent did it, the agent (or the tool) is responsible.
Responsibility can be delegated; accountability cannot.
Tasks and even role-work can be delegated; the answerability for the outcome cannot. A named human remains accountable for what ships — that floor is where seniority now lives.
Read the evidence → - 06
Conventional wisdom: Build provenance (SLSA) gives us an audit trail, so we’re covered.
Audit trails don’t create accountability.
Build provenance (SLSA) attests where and how an artefact was produced; it does not attribute the decision. Accountability needs a decision-and-attribution trail engineered above the pipeline.
Read the evidence →
How we arrived here
We did not start with AI. We started with architecture, product delivery, engineering and transformation. As AI accelerated, we saw organisations making a familiar mistake: treating a new technology challenge as though it were purely a tooling one.
The technology changed quickly. The underlying disciplines did not. Architecture still matters. Governance still matters. Decision quality still matters. The organisations succeeding with AI are not abandoning those disciplines — they are applying them to a new generation of tools. That observation became the foundation of Ivaaya.
We are deliberate about what we claim: the field is young, and nobody has decades of experience in agentic engineering. We bring hard-won discipline and apply emerging AI where it creates genuine leverage — honest about what we know, what we are testing and what we are still learning.
What we learned
- Larger context windows did not improve AI coding quality as much as curated engineering context did.
- The most expensive delivery failures were usually translation failures, not implementation failures.
- The governance controls worth keeping were the ones that could name the bad decision they prevented.
- Reliable connected systems failed at the seams — power, connectivity, provisioning — far more often than in the hardware.
These are lessons from building, not slides — the kind of thing that only shows up once you have shipped the work. They run through our insights, frameworks and capabilities.
Challenge the thinking
If this point of view resonates — or if you disagree with it — we would genuinely like to hear from you. The best conversations often start where the assumptions differ.
What does this make you think? →