Skip to content

AI and software studio in Boulder, Colorado

We build AI software.
For your business.

Zentered Studios builds products, integrations, and agent workflows with AI, and sets up your team to use them. You work directly with Patrick Heneise, who does the work, from the first call to the handoff.

Services

What we do for you

Set up your team, connect AI to your systems, or have us build the software. Not sure which fits? Start with a 15-minute call.

Help your team use AI

01

Claude Team setup

One day to roll out Claude Team with roles, permissions, and safety guardrails in place, plus a runbook for your administrator. Up to 50 seats. Fixed at $3,000; Anthropic subscriptions are separate. See the day.

02

AI training for your team

A four-week, in-person pilot in Boulder County. Your team puts AI to work on two tasks it does every week, measures the change, and writes its AI policy. Fixed at $6,500 for up to 10 people. See the four-week pilot.

Connect AI to your systems

03

MCP workshop

One day to decide whether a custom MCP server for your internal products is worth building, and what it should do. You leave with a ranked recommendation, including where not to build. Up to 12 people, in person or remote. Fixed at $2,000. See the day.

04

MCP servers: design, build, audit

An MCP server lets AI assistants such as Claude use your product. We design and build one, or we measure how agents use the one you have, find the tools they misuse, and ship the fixes. Details, or see mcpsignals.

Build your software

05

AI-native product development

We take your product from a scoped idea to working software, with the same AI-assisted process we use to build Roadbook. Patrick owns the architecture and the delivery. We build the whole product, or the first release and hand your team the process. Details, or see Roadbook.

06

Agent pipelines for your team

We turn the engineering work your team repeats every week, such as bug triage, code review, and regression checks, into AI agent workflows. They live in your repository, where your team can read, change, and run them. Details.

Decide where to start

07

AI consulting

Not sure where AI fits? Patrick reviews your workflows and systems, then tells you what to automate, what to keep human, and how to measure the result, before you commit to a build. Details.

Tell us what is slowing your team down

Tell Patrick what your team is trying to do, where the work stalls, and your deadline. We tell you whether we can help, and scope the next step at a fixed price.

Fixed price

One day, one price

Make a build decision or get your team set up. A defined scope, a working session with Patrick, and a handoff your team can use.

MCP workshop

$2,000

  • 1 day, in person or remote
  • Up to 12 people
  • For the people who own your internal products

Decide what is worth connecting before you pay to build it. Patrick works with the people who own your internal products to find where handoffs stall, assess whether MCP is the right fit, and draft tools for the strongest candidates. You leave with a recommendation you can take to your own engineers or another supplier. No code is written on the day.

The day

  1. Morning

    How MCP servers work and what makes a tool easy for an agent to use, with examples from servers we run.

  2. Late morning

    Walk through each internal product with the people who run it, and find where work stalls between products.

  3. Afternoon

    Decide where an MCP server makes sense, and sketch the tools it would offer and which of them write data.

  4. End of day

    Rank the candidates by value and effort. Recommend a first build, or explain why none is justified yet.

You leave with

  • ✓ A map of your internal products, workflow bottlenecks, and potential MCP integrations
  • ✓ A draft tool list for the top candidates
  • ✓ A ranked recommendation your team can act on, including where not to build. An optional implementation quote is separate

Pay by card or US bank transfer (ACH) through Stripe. Sales tax is added at checkout only if it is owed.

Claude Team setup

$3,000

  • 1 day, in person or remote
  • Up to 50 seats
  • You buy the seats from Anthropic

Roll out Claude with guardrails in place from day one. Patrick sets roles and connector permissions, turns your security and data policies into a guardrails skill, and adds an AI safety skill for untrusted content and unchecked output. Your administrator leaves with the setup documented and a runbook to maintain it. Bring your workspace owner and your security and data policies; buy your Anthropic seats separately.

The day

  1. Morning

    Set roles and permissions. Review every connector and turn off the access your team does not need.

  2. Late morning

    Write a guardrails skill from your security and data policies: what stays internal, what is never published, and what needs approval.

  3. Afternoon

    Write an AI safety skill: treat email, web pages and connector content as data, keep credentials out of chats, and check output before it reaches a customer.

  4. End of day

    Provision both skills organization-wide, check the setup, invite the team, and hand over the admin runbook.

You leave with

  • ✓ Roles and permissions set, and written down in an admin runbook
  • ✓ Guardrails and AI safety skills provisioned organization-wide and enabled by default
  • ✓ An invite and offboarding checklist for whoever manages seats

Pay by card or US bank transfer (ACH) through Stripe. Sales tax is added at checkout only if it is owed.

Claude Team setup

Guardrails in every conversation

A security policy in a separate document does not reach the conversation where someone asks Claude for help. We turn your rules into two skills your organization maintains centrally, with a clear owner and instructions for updating them.

Guardrails skill

Keep company data where it belongs

Claude will help publish an internal tool, paste source code into a public service, or share customer data if nobody tells it not to. The person asking may not know the rule either.

We translate your security and data policies into guidance for Claude: what counts as sensitive, where it may and may not go, and which actions need a human decision first. Your administrator can update the guidance centrally as the rules change.

What changes
  • ✓ Company-specific rules for internal tools, source code, and customer information
  • ✓ Explicit approval points before anything is published, deployed, or shared externally
  • ✓ An editable skill your team owns, provisioned across the organization

Skills guide model behavior; they do not enforce access controls or guarantee confidentiality. Owner-provisioned skills are enabled by default, but users can disable them. We document those limits in the handoff.

AI safety skill

Treat outside content as data, not instructions

Connectors bring email, documents and web pages into the conversation. Any of them can carry text written to steer Claude, and a confident answer can still be wrong.

We write guidance for how Claude handles what it reads and what it produces: outside content is data to summarize, never instructions to follow; credentials and secrets stay out of chats; and facts, figures and customer-facing drafts get a human check before they leave the company.

What changes
  • ✓ Instructions to flag, not follow, requests found inside emails, documents and web pages
  • ✓ A rule to keep passwords, API keys and access tokens out of conversations
  • ✓ Review points before output reaches a customer or a public channel

This reduces risk; it does not remove it. A skill cannot block a prompt injection or catch every wrong answer. We document those limits in the handoff.

See the Claude Team setup day

Payment
Paid in full before the day, with an invoice for your records. Remote days cost the same as in person.
Travel
Included within Boulder County. Anywhere else is agreed in writing before we book. Out of state, we agree a travel budget in writing first.
Cancellation
Full refund if you cancel 7 or more days before the date. Less than 7 days: no refund, and you can move the date once, within 60 days. Full terms
Bigger teams, more days
More days, and teams over the size limit, are priced on request.

Send us the team size and the number of days, and we send back a fixed quote.

Request a quote

Work

What we have built

Products we build and run, an open-source library, and the AI rollout Patrick led before the studio. Each case says where it started, what the work was, and where it stands today.

Nevados, before Zentered Studios

Claude and an agent platform for 70+ people

Starting point
A utility-scale solar tracker manufacturer with engineering, supply chain, sales, and operations teams, and no company AI policy yet.
The work
  • ✓ Rollout of Claude and an internal agent platform across engineering, supply chain, sales, and operations
  • ✓ Integrations with HubSpot, Monday.com, and Salesforce for sales and product management
  • ✓ The company's first AI policy and an enablement program
  • ✓ A guardrail skill, provisioned company-wide after Claude suggested publishing an internal tool to Vercel
Result
More than 70 people in four departments worked with Claude and the agent platform. Patrick led the work as Director of Software, then Director of AI & Platform.

Our product

Roadbook: an EV road trip planner

Starting point
EV drivers planning a long trip need chargers, campgrounds, and parks on one route, across the US, Canada, and Mexico.
The work
  • ✓ A web app that plans each leg around chargers, with drive times, fees, and typical weather
  • ✓ Places from OpenStreetMap, NPS, and Open Charge Map, and shared trip planning for a group
  • ✓ An MCP server, so AI assistants can plan with it, instrumented with mcpsignals
  • ✓ Built with the agent pipeline we sell: spec, plan, isolated worktree, review, and browser verification
Result
Live at roadbook.us. The same process runs our client builds.

Open source, MIT

mcpsignals: see what agents do with your MCP server

Starting point
Teams ship MCP servers without knowing which tools agents call, which calls fail, or which responses use up a session.
The work
  • ✓ A drop-in library for Node and Python
  • ✓ Records tool calls, errors, latency, payload size, and the agent's stated intent
  • ✓ Writes to a database you own: Postgres, BigQuery, Cloudflare D1, or an OpenTelemetry collector
Result
Records agent calls in Roadbook. It is the method behind our MCP server audits.

This website

A homepage AI agents can use

Starting point
An AI assistant acting for a visitor has to scrape a page to learn what a company sells or how to reach it.
The work
  • ✓ Thirteen WebMCP tools: read services, products, our work, training, case studies, testimonials, and page Markdown, send a message, request a quote, check call times, book a call
  • ✓ An RFC 9727 API catalog, an OpenAPI 3.1 spec, llms.txt, and a Markdown version of every page
  • ✓ Every tool call logged to Cloudflare D1 in the mcpsignals event shape
Result
In a browser that supports WebMCP, an agent can request a quote or book a call without parsing the page.

More we have built

GitMorning

AI-generated standups from real GitHub activity, with no self-reporting. Not open for sign-ups yet.

Join the interest list

Rivian MCP

An open-source, read-only MCP server and CLI for Rivian vehicles: battery, range, charging, and software updates. It never sends vehicle commands.

View on GitHub

Want something like this built for your business?

We build for clients with the same process and tools you see above. Tell us what you want to build, and we scope the first step at a fixed price.

Book a 15-minute call

Blog

Building in public

Notes on AI-assisted product development from Boulder, Colorado.

Ten Tickets or One: Cutting Setup Cost in Our Agent Loop

Our agent pipeline worked in small batches without cutting setup time, so even a tiny ticket used millions of tokens across its delivery cycle. What we measured moving from GitHub Projects to Milestones, what got worse, and the changes we make next.

Your CMS Is Now a Slack Channel

We wired Claude Tag to a marketing site built on Astro, SolidJS, and plain Markdown files. Content managers ask for changes in Slack, Claude opens a pull request, Cloudflare builds a preview, and the change ships when someone approves it on GitHub - no CMS, no admin login, no deploy button.

All posts

About

Who you work with

Zentered Studios LLC is an independent AI and software development studio in Boulder, Colorado, run by Patrick Heneise. Agents do the implementation. You talk to and work with a person, from the first call to the handoff.

Patrick spent four years building software at Nevados, a utility-scale solar tracker manufacturer, as Director of Software and then Director of AI & Platform. His work covered the software, the integrations, and the people expected to use them. Today, he brings that experience to client engagements and the products Zentered Studios builds and operates.

70+
people on Claude and our agent platform at Nevados
4
departments involved in the rollout
4 years
building software at Nevados
Patrick Heneise
“The interesting part was never the model. It was adoption, which means sitting with a team and learning which spreadsheet they actually open on a Monday.”
Patrick Heneise, Founder, Zentered Studios

Co-founder of BoulderJS. Writing software since 2000.

Start a project

Tell us what you want to build

A 15-minute call is enough to tell whether we can help. If we can, we scope the next step at a fixed price.