Most agencies will not put a number on a page. We understand why. Ranges get quoted back out of context, and the honest answer really does start with "it depends."
It still leaves you unable to plan. So here is what we charge for AI agent work, what each band covers, and where a packaged Sprint ends.
Short version: most projects land between $1,500 and $22,000 as a one-time setup. Simple Agent Sprints are $1,500–$4,000. Medium builds are $6,000–$11,000. Advanced custom AI is $14,000 and up. Running costs usually add about 5–15% of setup per year. Well-scoped work often pays for itself in two to six months.
Cost bands at a glance
| Band | Setup | Typical timeline | What you get |
|---|---|---|---|
| Simple | $1,500–$4,000 | 1–3 weeks | One clear workflow, fixed-price Agent Sprint |
| Medium | $6,000–$11,000 | 4–8 weeks | Deeper logic, stronger evaluation, systems of record |
| Advanced | $14,000+ | 8+ weeks | Multi-agent, product-embedded, or combined product work |
That overall $1,500–$22,000 range covers most first releases we see. Larger platforms that wrap agents in a full product surface can go beyond it and get scoped like any other custom software engagement.
Who these engagements are for
Any company with recurring work that spans tools and eats real hours. Industry does not matter. Company size does not matter. If people on your team keep doing the same multi-step job by hand, and you can describe what a correct result looks like, that is the kind of workflow we automate.
We connect to the systems you already run: CRM, support, ERP, email, sheets, HRIS, industry tools, whatever the job needs, as long as we can reach them with a stable API or export path. The bands above are just where planning starts.
Simple builds: Agent Sprint tiers ($1,500–$4,000)
Inside the simple band we sell fixed-price Agent Sprints, paid upfront. Complexity is set by workflow depth and whether the agent carries memory that evolves across runs.
Focused: $1,500
Delivery: 5 to 7 business days. One revision round. Two-week post-launch tuning window.
Single-pass workflow: the agent reads, drafts or flags, and stops. No persistent memory beyond basic duplicate avoidance.
This is the shape of a Friday status-update agent or a silent churn-signal watcher. One job, a short path from input to output, and a person still decides what happens next.
Connected: $2,500
Delivery: 7 to 9 business days. Two revision rounds. Three-week post-launch tuning window.
Multiple sources feed one decision, with a scoring or ranking step before the agent drafts.
This is the shape of a lead-gen agent that scans Reddit, the open web, LinkedIn, and X, ranks leads, and posts with reasoning plus a draft reply. The interesting work is the ranking and the judgment call, not just the fetch.
Full-Scope: $4,000
Delivery: 10 to 12 business days. Two revision rounds. Four-week post-launch tuning window.
Multi-step workflow with layered decision logic, plus memory that carries and updates across runs.
This is the ceiling of the Agent Sprint ladder. The right call when a Connected agent would feel stretched.
Medium builds: $6,000–$11,000
Medium work is for projects that no longer fit a packaged Sprint. Typical signals:
- Several workflows that have to share context
- Agents that write into systems of record and need stronger evaluation and human review
- Workflows where the product decision is still being made alongside the build
- A first agent plus light surrounding UI, ops tooling, or admin review queues
These engagements are scoped after discovery and priced as fixed scope. Timelines usually land in the 4 to 8 week range.
Advanced custom AI: $14,000 and up
Advanced work is product work that happens to include agents. Examples:
- Multiple agents that coordinate with each other
- Automation embedded inside a customer-facing product
- Engagements that combine agent development with product strategy, design, and broader engineering
Most well-bounded first releases in this band still land under about $22,000. Larger platforms are quoted like any other custom software project.
What actually moves a project between bands
Workflow steps. A single-pass "read, draft, stop" agent stays in the simple band. Multi-source gather-score-rank-draft work can still be a Sprint. Layered decision logic with state across runs sits at the top of simple or tips into medium.
Memory. An agent that starts fresh each run is simpler than one that has to remember what it already flagged, who it already contacted, or how a score moved since last week.
The cost of being wrong. An agent that drafts a status update can be wrong occasionally, because a human reads it before it goes anywhere. An agent that writes into a system of record cannot. Higher stakes raise evaluation and review effort, which can move a project from simple into medium.
Product surface. Auth, dashboards, billing, or a customer-facing product around the agent is advanced work, not a Sprint.
What is included, and what is still open
Every Agent Sprint includes:
- Fixed scope and fixed price, paid upfront
- Delivery inside the business-day window for that tier
- The revision rounds listed above
- A post-launch tuning window (2, 3, or 4 weeks depending on tier) so we can adjust prompts, thresholds, and edge cases against real usage
Medium and advanced engagements include a fixed scope after discovery, the same evaluation and human-review bar we hold all agent work to, and documentation for handoff.
What support looks like after any included tuning window is still an open decision on our side. Do not assume a retainer is included, and do not assume there is none. If ongoing maintenance matters to your decision, ask us before you buy.
Ongoing model and infrastructure costs sit with you after launch. For most business workflows that means model API usage that scales with run volume, plus light hosting. As a planning rule of thumb, expect running costs around 5–15% of setup per year. We design for that during the build by choosing right-sized models per step and keeping context tight.
When the agent replaces recurring manual work, well-scoped projects often pay for themselves in two to six months.
When you should not buy an agent project
The cheapest agent project is the one you do not run.
- An off-the-shelf tool already does it. Standard support triage, meeting notes, and generic sales outreach have products. Custom is worth it when the workflow is specific to how your team works.
- The workflow is not stable yet. If the process changes every month, you will be paying to rebuild the agent every month. Stabilize the process first.
- Nobody can describe a correct answer. If your own experts disagree about what the right output is, an agent cannot be evaluated, and an agent that cannot be evaluated cannot be trusted.
- You need a platform, not an agent. That is advanced custom work. Price and timeline should reflect the product, not a stretched Sprint.
Questions to ask any firm you are evaluating
- How will we measure whether the agent is right?
- What happens when it is wrong? Where is the human checkpoint?
- What are the projected monthly model costs at our volume?
- Who owns the code and the prompts when we are done?
- What does support look like after the post-launch window, and what does it cost?
- Can you show a workflow like ours that you have shipped?
If a firm cannot answer the first two clearly, the price is not the thing to worry about.
Where we land
Most AI agent projects with us cost between $1,500 and $22,000 as a one-time setup. Simple Agent Sprints are $1,500, $2,500, or $4,000. Medium builds typically land at $6,000–$11,000. Advanced custom AI starts around $14,000. Plan for running costs of about 5–15% of setup per year, and for well-scoped work to pay for itself in two to six months.
If you want the shape of the work in more detail, our AI development page covers approach and what we build. The workspace problems series walks through specific workflows and the agent we would design for each.