Agent workflow map
User goals, triggers, tool actions, handoff states, limits, fallback paths and where a human should stay in control.
Service · AI Agents & Bots
Custom AI agents built for workflows that need more than a chat box.
99 Francs designs and builds AI assistants, agent workflows, support bots, internal copilots and LLM product surfaces that are understandable, controlled and ready for production.
A simple single-workflow agent from $2,500; typical custom agents run $5,000–15,000.
150+
shipped projects
$32M+
raised by clients
9,000+
tasks delivered
Direct answer
99 Francs designs and builds AI agents around real product and workflow needs.
This service is for teams that need a custom AI assistant, workflow automation, chatbot, internal copilot, RAG flow or AI product designed, integrated, tested and launched by one senior team.
Deliverables
What the AI agent scope can include.
The work focuses on the visible experience and product logic: what the agent knows, what it can do, how users control it and how the workflow should behave.
Chat and task UX
Conversation structure, prompt inputs, output states, review steps, confidence cues, error states and reusable UI patterns.
Assistant product surface
High-fidelity screens for copilots, support bots, internal assistants, AI search, dashboards and launch demos.
Production build and integration
Agent logic, provider fallbacks, React or Next.js product surfaces, APIs, business-system integrations, tests and deployment.
Best fit
Use this when the agent has to work inside a real business flow.
The idea is promising, but the workflow is vague
We turn the agent concept into screens, states, boundaries and practical actions before engineering starts.
A chatbot alone is not enough
We design the surrounding product experience: onboarding, inputs, citations, escalation, approvals, history and reporting.
The demo needs to sell the product
We package the AI workflow into a polished MVP surface or launch demo that makes the value visible quickly.
Trust and control are unclear
We make limits, confidence, human review and next actions visible so users know when to rely on the agent.
Process
Prove the agent before asking users — or your business — to trust it.
The process keeps AI capability, user control, product clarity and production reliability connected from the first workflow to launch.
Define the job
We identify the workflow, user roles, data sources, allowed actions, risk points and what the agent should not do.
Map states and control
We design input, response, review, escalation, memory, error and handoff states before polishing the interface.
Create the product surface
We produce the screens and interaction system for the assistant, bot, copilot, workflow or product demo.
Build, test and launch
We implement the agent and its product surface, connect the required systems, test the action map and validate it safely before production.
Related work
Relevant work includes AI products, search tools and workflow-heavy interfaces.
AI agents need interface clarity as much as model capability. These projects show product surfaces where complex systems had to become usable.
Algo trading
Polymarket Trading System algo & market-making
Custom automated trading and market-making system on Polymarket. A Python quant engine with a low-latency Rust execution core, a React dashboard, backtesting, and a safe virtual-to-real rollout.
Mobile app
Digsdy real estate
AI-powered real estate platform that understands user queries in chat, analyzes preferences, and delivers relevant property listings in real time. Built as an MVP in Webflow.
Guides
Further reading
Process7 min read
Where We Use AI in Client Work, and Where We Don't
Plenty of business owners now read AI as a sign that a vendor cut corners. The real problem is AI output nobody checked. Here is what we hand to an agent, what stays with people, and how one real WordPress-to-Webflow migration split.
Research11 min read
How to Design AI Agents Users Trust: UX Patterns for Control, Confidence, and Recovery
A field guide to designing AI agents people actually trust — six UX patterns for control, autonomy, uncertainty, memory, voice, and failure, each drawn from agents we have shipped.
Process10 min read
How We Build AI Agents: Our AI Agent Development Process, From Feasibility to Production
How we design and build AI agents that survive production, not just the demo — feasibility proof, behavior prototypes, tested action maps, provider fallbacks, and an outcome-based model where you pay only when the agent works.
Buyer guides13 min read
AI Agent Development Cost in New Zealand: 2026 Pricing Guide
What a production AI agent costs in New Zealand in 2026 — public NZ market benchmarks, realistic custom-build ranges, ongoing costs and a lower-risk way to pay only after each phase works.
AI Agents FAQ
Make the workflow clear before turning it into automation.
AI agent and bot design can include workflow mapping, chat UX, assistant screens, prompt inputs, output states, review steps, fallback paths, escalation logic and build-ready interaction notes.
Yes. 99 Francs can take an agent from feasibility and workflow design through the product surface, model and tool integrations, testing and production deployment. The exact backend and infrastructure scope is defined before each phase starts.
No. The service can cover internal copilots, AI search, sales assistants, operations agents, workflow automation, onboarding bots and product demos.
Start with design when the workflow, user control, trust model, screen states or business value are unclear. Development is easier once those decisions are visible.
A simple single-workflow agent starts from $2,500. Typical custom agents — with tool integrations, guardrails and a review flow — run $5,000–15,000, depending on scope, data sources and how much human oversight is built in. Every project runs in fixed-price phases with no upfront payment. See how we work for the full model.
For New Zealand clients, comparable custom AI-agent work is priced at NZ$1,750–17,500, with no GST added on top — 30% below the public NZ$2,500–25,000 local range. Advanced systems are quoted in phases, and other markets are priced in the client's preferred currency after scope and acceptance criteria are agreed.
A simple, single-workflow agent usually takes one to two weeks. A more complex agent — with integrations, tests, approvals and production deployment — typically takes two to six weeks. Larger multi-system agents are split into independent phases, and each phase is paid only after its working result is accepted.
No. Custom work is split into fixed-scope phases. You pay only after a phase is demonstrated working and accepted; if it still misses the agreed outcome after revisions, you owe nothing for that phase.