The Multi-Agent AI Platform We Built to Fix Our Own Sales Bottleneck
.jpg){kind=small}
Written by
Subject
Published on
Over the past two years we rebuilt our own development process around AI-assisted coding — today over 95% of our own code is AI-assisted, up from close to zero. Delivery got faster. The work that happens before delivery did not: understanding what a client actually needs, structuring it, and pricing it. Scoping and quoting became the bottleneck standing between a good idea and a working build. So we built ChatStack, a multi-agent AI platform, to fix it — for ourselves first.
Turning a client conversation into a structured, priced brief has always taken weeks: a discovery call, a written brief, several rounds of clarifying questions, and a manually built estimate. None of that gets faster just because the engineering behind it does. For a business that had just cut its own build time dramatically, a two-month sales cycle looked less like due diligence and more like friction.
ChatStack replaces the discovery call and the written brief with a single structured session. An interview-driven, four-agent workflow asks the questions a good business analyst would ask, in the order a good business analyst would ask them, and turns the answers into a Product Requirements Document — user stories, functional and non-functional requirements, and a technical specification — alongside a fixed-price, story-point-based cost estimate. The whole session takes 10 to 20 minutes.
A structured discovery interview, run by a coordinated set of specialised AI agents, replaces the open-ended briefing call.
Requirements are delivered as structured CSV or JSON, built to drop straight into Cursor, Claude Code, Windsurf, and other MCP-aware IDEs.
A transparent, story-point-based pricing model, rather than a black-box number, with AI-assisted delivery efficiency built into the underlying calculation.
Beyond the human chat interface, ChatStack exposes an authenticated API, documented with a published OpenAPI 3.1 specification, so autonomous AI agents can submit and retrieve requirements directly.
Each session runs 10 to 20 minutes: short enough to hold a prospective client's attention, long enough to extract real detail.
Compressing weeks of scoping into one short session risked shallow output. We solved it by structuring the interview around fixed, agent-led stages rather than an open-ended chat, so nothing material gets skipped.
An estimate a client can act on has to be defensible, not just fast. We solved it with a transparent story-point model, rather than a single unexplained number.
We use ChatStack ourselves — it is the tool behind our own “Get a Quote” button. It cut our own sales lead-to-conversion timeline from two months to three weeks. Once other teams saw what it did for our own pipeline, we opened it up so they could use it too.
We are extending ChatStack’s agent workflow, broadening API access for autonomous agents that integrate over the Model Context Protocol, and building a full-service tier for teams that want us to execute the resulting brief as well as generate it.
ChatStack is proof that we practise what we advise. When a bottleneck showed up in our own business, we did not write a strategy document about it — we built the multi-agent system that removed it, then measured the result. It is the same process we now bring to enterprise clients through our AI Consulting practice.
ChatStack turns a conversation into a structured requirements document and a fixed-price estimate in one 10-to-20 minute session — the same process we now bring to enterprise clients through our AI Consulting practice.
AI-powered car damage detection, built with Tcheck on a model trained on 100+ cars.
A compassionate AI chatbot combatting gender-based violence. GRIT / Zuzi AI Award Winner, 2024.
A personal security app fighting gender-based violence in South Africa. Winner, MTN/FNB App of the Year, 2022, 2023 and 2024.
A dream interpretation, social media app. People record their dreams in various media formats and engage on dreams with other people and experts.
