Back to all work
ResearchResearch

Designing an AI-Assisted Project Delivery System

AI agents for task creation, structured workflows, human-in-the-loop review, and sprint execution for consulting and software delivery.

Type
Research
Status
Research
Tags
AI · Agents · Delivery
/ 01

Problem

Consulting and software delivery move slowly because coordination is manual. Task creation, sprint planning, status reporting, quality checks — each engagement rebuilds these from scratch. AI helps in pockets, but the gains don't compound across projects.

/ 02

Challenge

How do you thread AI agents into delivery workflows without removing human judgment at the gates that matter — prioritisation, scope decisions, quality sign-off?

/ 03

Proposed solution

An operating system where AI agents propose tasks, plans, and quality checks, while humans approve at defined gates. The system records delivery patterns across engagements so the next project starts further down the learning curve.

/ 04

Technology used

TypeScriptAgent frameworksStructured plan schemasProject state machineAudit trail / event log
/ 05

Business value

Not yet measured. Working hypothesis: 30–50% reduction in coordination overhead for a typical engagement, with quality maintained or improved at the human-review gates.

/ 06

Current status

Research. Currently designing the gate model and the agent contract.

/ 07

Lessons learned

AI-assisted works at task generation and quality-check stages. AI-driven does not yet work at the prioritisation stage — humans still own that call, and pretending otherwise produces plans that look right but feel wrong. Internal research finding.

/ 08

Future roadmap

Cross-engagement learning, automated quality gates, pluggable agent providers, client-facing read-only views.

Have a problem that looks like one of these?