Private operator
An assistant that lives in your channels with only the skills and access you approve. Good first agent for a small team.
AI agent builds
Not a demo.
A teammate.
Most agent projects stall after the demo, because the demo was never the hard part. We start from a proven agent and spend the hours on the part that makes it useful: your workflow, your data, your permissions, and the failure cases.
What the hours go onHow we budget it
Shaping the workflow25%
Connecting your systems30%
Permissions and safety20%
Testing against real cases15%
The model and prompts10%
This is our planned use of the hours, not an average of past builds. The model and the prompt are the small part. Connecting it to real systems and proving it behaves is where the time goes.
01Pick a starting point
02Kickoff scopes the job
03We connect and harden it
04You keep the code
Starting pointsPick the closest one and we adapt it
Start from
something that runs.
Each of these already works. What it does not know is your business. That is what you are paying us to teach it.
Lead and CRM agent
Qualifies, enriches, routes, updates the record, and follows up. Sits on top of Keap, HubSpot, or your own pipeline.
Knowledge and support
Answers from sources you approve, says so when it does not know, and hands off to a person with the thread intact.
Something of your own
A job none of the above fits. We scope it in the kickoff and build it on the same footing as the rest.
The real workWhere agent projects usually die
The demo is easy.
The rest is the job.
Five things separate an agent you show people from one your team actually uses. We build all five, and we tell you before we start which are hard in your setup.
01
The job
Deciding what the agent owns end to end, and what it must hand to a person. Vague scope is why agents get abandoned.
02
The connections
Real reads and writes into your CRM, database, inbox, or ticket system, with the edge cases your data actually contains.
03
The limits
What it may see, what it may change, what needs approval first. Logged, scoped, and reversible.
04
The failure cases
What happens when it does not know, when a system is down, or when the answer would be wrong. Handled, not hidden.
05
The proof
Run against real examples before launch, so you see the hit rate rather than taking our word for it.
The price
Three ways in.
Priced before you start.
Senior team rate$100/hranything beyond a package
Prebuilt Agent Launch
1agent, live
A proven agent starting point, deployed securely and configured for your business.
Best for: A focused assistant or operator with one channel and a clear first job.
$750prepaid
- One agreed agent starter
- Model provider and BYOK setup
- One communication channel
- Security baseline and handoff
- 7 days of defect support
Agent + Custom Workflow
1workflow, wired in
A working agent connected to one real workflow and one business system.
Best for: A team that already knows the workflow it wants to improve.
$1,500$750 deposit
- Everything in the prebuilt launch
- One custom business workflow
- One CRM, email, Slack, or API integration
- Prompt, skill, and workflow testing
- 30 days of defect support
Custom Agent Discovery
5hours of scoping
Architecture and a build plan for a custom, higher-stakes, or multi-system agent.
Best for: A complex idea that needs a safe architecture and a clear estimate first.
$5005 hours prepaid
- Workflow and risk mapping
- Integration and data audit
- Recommended architecture
- Deployment and security plan
- Fixed estimate or build roadmap
Not sure which one? Start with the prebuilt launch. You will know inside the first week whether the agent earns its place, and everything we set up carries straight into the workflow build.
Hosting and model usage are billed to you directly.
Questions
Before
we start.
Which model do you use?
Whichever fits the job and your constraints. We recommend one after the kickoff, and we build so you can change it later without rewriting the agent.
Does our data go into a model's training?
Not through anything we set up. We use API access with training turned off, keep your data in your own systems where we can, and tell you plainly what leaves your network.
What does it cost to run each month?
Hosting and model usage, billed to you directly. We estimate the monthly range during the kickoff so you can decide before we build.
What if it gives a wrong answer?
We build the agent to say when it does not know and to hand off to a person. Before launch we run it against real examples and show you where it fails.
Can it act, or only answer?
It can act, with limits you set. Writes to your systems are scoped, logged, and reversible, and anything risky asks a person first.
Who owns what you build?
You do. The code, the prompts, and the configuration are yours, in your repositories and your accounts.
Practical guidesUseful before you hire
Know what to ask before you start.
All guidesPractical automation · 4 min read
Which part of your business should you automate first?
Still copying customer details between apps or chasing the same updates? Learn how to pick one task, decide whether it needs AI, and check that it saves time.
A practical guide from WeNeedADev