AI-First SDLC - 28Stone Website Preview

LEVEL 1Traditional Agile DevelopmentThe universal starting point. A full team of nine: dev lead, four developers (two UI, two backend), manual tester, automation tester, business analyst, and project manager. Every line of code is human-written.

THROUGHPUTThe baselineThe industry standard for over two decades. Predictable, auditable, fully human-driven — and no longer competitive when measured against what a restructured team can deliver.

C-Suite

Predictable costs, predictable timelines, predictable output. Every metric is well-understood. The challenge is that your competitors are now delivering the same scope in a fraction of the calendar time.

Delivery Team

Full traditional team of nine. Two-week sprint cadence with sequential phases — backend waits for frontend, testing waits for both. Reliable process, but the calendar time is dominated by handoffs, not implementation.

Team Structure

Feature Team

Team of 9 professionals dedicated to the specific feature development. A structure that has served us well for the last 2 decades.

UI Dev
UI Dev
BE Dev

BE Dev
Manual Tester
Automation Tester

Dev Lead
Business Analyst
Project Manager

Sprint StructureThe standard two-week cadence. Work flows through stages in order: plan, build, test, integrate, ship. Calendar time is dominated by handoffs between phases.

LEVEL 2AI-Assisted DevelopmentMost organizations have development teams operating at this level in 2026. Individuals working smarter, same team, same cost, modest gains.

THROUGHPUTEveryone has CopilotIndividual productivity up 10–20%. Organizational throughput barely moves. Same team, same sprint, same bottlenecks — just faster typing.

C-Suite

License costs are minimal. Productivity gains are real but small — 10–20% individual, nearly invisible in delivery timelines. This is where most enterprise AI investment sits today: low risk, low disruption, low return.

Delivery Team

Same nine people, same two-week sprint. Developers use Copilot, BAs draft specs with ChatGPT, testers generate test cases faster. Everyone is individually more productive. The sequential workflow — and its bottlenecks — remains identical.

Team Structure

Feature Team

Team of 9 professionals dedicated to the specific feature development.Same team. Same cost. Modest gains. This is not structural change — it is individuals working smarter. Further improvement requires a different model.

UI Dev
UI Dev
BE Dev

BE Dev
Manual Tester
Automation Tester

Dev Lead
Business Analyst
Project Manager

Sprint StructureSame two-week sprint, same sequential flow. Individual tasks complete faster, but the phases still run in order. The sprint structure hasn't absorbed the productivity gains.

LEVEL 3Experimental / Early RestructuringBetween Level 2 and Level 3, teams see real gains — enough to feel good about, not enough to change the economics.

THROUGHPUTFirst real gains First headcount reduction. The team shrinks by two. A fullstack developer uses AI more aggressively. Throughput is up 20–40% — enough to satisfy a quarterly review, not enough to change the economics. It feels like progress, but the fundamental structure is still intact.

C-Suite

The organization has discovered AI tooling organically and made the first structural changes. Two headcount removed. Some delivery efficiency gained. But the sprint is still two weeks and the gains are inconsistent across the team. This is the most common state we find on capital markets engagements — and the hardest to move past without a deliberate model.

Delivery Team

You've shrunk the team from 9 to 7. You've saved two headcount. Your developers are using agents for unit test generation and code review automation. But the sprint is still two weeks, and uneven AI adoption creates delivery inconsistency. This is where most organizations plateau.

Team Structure

Feature Team

Between Level 2 and Level 3, teams shrink to 7 people and see real gains — enough to feel good about, not enough to strategically change the approach to software development.

Dev Lead
UI Dev
BE Dev

Project Manager
Manual Tester
Automation Tester

Business Analyst

Sprint StructureStill two-week sprints. Some overlap between phases begins as the fullstack developer works across streams. But planning, testing, and integration remain largely sequential.

LEVEL 4Agentic Development28Stone's main production model. 3–5× throughput at half the team cost. This is the structural leap. Delivery economics fundamentally change — a shared architect, BA, and PM support multiple three-person teams, each running parallel AI streams at 3–5× velocity. This is where compounding begins.

THROUGHPUT3–5× delivery velocityOne-week sprints. Three dedicated engineers replace a nine-person team. Agents generate code in parallel across all streams — humans architect, verify, and own the compliance boundary.

C-Suite

This is 28Stone's operating model on active capital markets engagements. A three-person feature team delivering at three to five times the velocity of a conventional team. Sprint cadence compresses to one week. At this level, the ROI conversation shifts from 'should we try AI?' to 'how many feature teams can we run in parallel?'

Delivery Team

One-week sprints. Monday alignment. Friday demo. AI handles implementation scaffolding, test automation, and code review first-pass. The human team does integration, judgment calls, and compliance-sensitive code. Three to five times the throughput at roughly half the team cost. The Architect role (evolved from Dev Lead) serves 2–3 feature teams simultaneously.

Team Structure

Feature Team

Dedicated professionals

UI Dev
BE Dev
Automation Tester

Shared Team

Working on 2–3 projects

Architect (from Dev lead)
Business Analyst
Project Manager

Sprint StructureOne-week compressed sprint. Human alignment on Monday, Tuesday–Thursday parallel execution across frontend, backend, and testing simultaneously. Thursday integration. Friday demo and deploy.

LEVEL 5Agent OrchestratedThe orchestrator replaces the team. Throughput gets to ~8× compared to traditional agile development. A single engineer defines intent, reviews results across multiple streams, and personally owns every compliance-critical code path. Impressive gains, but it demands a rare set of skills that can't be shortcut. Industry knowledge and deep understanding of specifics plays a more significant role than ever.

THROUGHPUTAn orchestrator replaces the teamOne senior engineer orchestrates four or more AI agents simultaneously. Throughput reaches ~8× at roughly 28% of traditional team cost. The human role shifts entirely to governance, exceptions, and architecture.

C-Suite

One senior engineer manages four or more AI agents running concurrently. The human role shifts entirely to governance — defining intent, reviewing exceptions, approving architectural decisions. At roughly 8× the output at roughly 28% of traditional team cost, this is the inflection point where delivery economics change category.

Delivery Team

The Technical Orchestrator writes almost no production code. They define agent tasks, review outputs, approve merges, and personally own the compliance-sensitive code paths that regulators care about. Authentication, trade execution logic, regulatory reporting, DB migrations — those are human-owned domains. A named human reviews, approves, and is accountable for every line. Everything else is agent-delegated.

Team Structure

Feature Team

Dedicated team of one senior engineer and a swarm of AI agents

AI Agents Orchestrator

Shared Team

Across 2–3 teams

Architect (from Dev lead)
Business Analyst
Project Manager

Sprint StructureOne-week sprint, further compressed. The orchestrator launches agent swarms Monday morning. By Wednesday, integrated code is under review. Thursday–Friday: validation, compliance review, deploy.

LEVEL 6Intent-Driven DevelopmentThis is a theoretical level included for strategic awareness. Not production-ready for regulated capital markets. The compliance boundary assumptions of Levels 4–5 do not hold at this level.

THROUGHPUTTwenty agents, one overseerAgents running in parallel. Full autonomy. Persistent memory across agent swarms. Continuous delivery replaces the sprint model entirely. The governance frameworks for this level of delegation do not yet exist.

C-Suite

20 to 30 agents running in parallel, backed by a persistent memory layer. The human Overseer sets strategic intent and reviews exception cases. Design and governance become your bottlenecks — not implementation. We do not recommend this operating model for capital markets systems under current regulatory frameworks.

Delivery Team

There is no sprint. There is no handoff. There is intent, execution, and exception review. We include this level for completeness and because your competitors will be asking about it. The compliance boundary that works at Levels 4–5 does not hold here. Level 5 is our recommended ceiling for regulated environments.

Team Structure

Feature Team

Dedicated team of one senior engineer and a swarm of AI agents

AI Agents Overseer

Shared Team

Across 2–3 teams

Architect (from Dev lead)
Business Analyst
Project Manager

Sprint StructureNo sprints. Continuous delivery. Agents execute against strategic intent, delivering increments continuously. The human overseer reviews exceptions and sets direction — the concept of a fixed sprint cycle no longer applies.

AI sharpens the compliance boundary. Maturity is also knowing when NOT to use AI.

Auth & Authorization

Trade Execution Logic

Regulatory Reporting

DB Migrations

Automated Testing

UI Implementation

Six Levels of AI-First SDLC: From Copilot to Orchestrator

C-Suite

Delivery Team

Feature Team

C-Suite

Delivery Team

Feature Team

C-Suite

Delivery Team

Feature Team

C-Suite

Delivery Team

Feature Team

Shared Team

C-Suite

Delivery Team

Feature Team

Shared Team

C-Suite

Delivery Team

Feature Team

Shared Team