MaverickX
Web development for MaverickX, a Y Combinator–backed company presenting chemistry across energy, mining and water treatment.
CREATE SOMETHING / AI-native technical support
Keep your tools working.
We diagnose problems, engineer improvements, and help your team use the result.
Support from $900/month · Agreed scope. See membership and costs ↗
The lesson stays with your team.
A checked handoff, instructions, and recovery checks preserve the context.
- Problem
- Approval
- Working system
- Captured lesson
Still view / the whole story Silent · no motion
Canon service guide · local prototype / scripted answers
How support works
From problem to checked handoff.
Follow one form that stops reaching a team. Inspect the decision, the checks and the instructions that stay with them.
Synthetic example · no connected systems, messages or bookings
- 01ProblemTrace the break
- 02ApprovalTeam owner decides
- 03CheckSuccess + failure
- 04HandoffInstructions + recovery
The same request stays in view as you inspect each part. These illustrations explain the method; they do not show a live result.
Problem / illustrative example
Find the broken handoff.
A form accepts a request, but the team never sees it. First, trace one synthetic request from the form to its destination.
Diagnosis comes before a proposed change. No system is connected here.
What we would inspect
- Expected
- One valid request reaches the team queue.
- Observed in this example
- The form shows success; the queue has no matching record.
- Still unknown
- Whether routing, permissions or delivery caused the break.
Approval / illustrative example
Agree the change before building.
The team owner reviews a bounded proposal: repair the handoff, check one agreed example, and keep a recovery path.
Inspecting this example does not approve work or grant an agent authority.
Illustrative scope record
- Owner
- The team owner decides whether the proposed change may proceed.
- Proposed work
- Repair one form-to-queue handoff and document its configuration.
- Outside the scope
- A new CRM, broader automation and production incident response.
Check / illustrative example
Check the failure case, too.
A working success message is not enough. Compare the destination record with the request, then check how an invalid request is handled.
These are proposed checks, not measured results or a customer success claim.
Illustrative acceptance checks
- Success case
- One valid synthetic request creates one matching queue record.
- Failure case
- An invalid request names the missing field and creates no queue record.
- Recovery
- If the outcome is unknown, reconcile the record before trying again.
Handoff / illustrative example
Keep the instructions and the next step.
The handoff connects the agreed change, checks, owner and recovery instructions. Your team can see what was done and what still needs attention.
Prepare the fixed sample below to see a complete local handoff. Nothing is sent or booked.
Illustrative handoff contents
- Keep
- The agreed deliverable, configuration notes and acceptance checks.
- Know
- Who owns the workflow and how to recover or ask for help.
- Next decision
- Review the scope for the next improvement; launch is agreed separately.
Try the handoff
A small, complete example.
Choose a fictional destination, review the scope and keep a sample receipt here. This does not request support or approve real work.
Interactive sample preparation needs JavaScript. All four records above remain readable without it.
Example handoff: form → team inbox; owner reviews one repair; check valid and invalid requests; retain instructions and recovery.
What we’ve built
From the agent’s tools to the customer’s screen.
We build the software around the workflow: the tools an agent calls, the checks that protect a decision, and the interface a person uses. Here are seven examples you can inspect.
- 01Developer tooling
Ground
A code-analysis tool that checks the evidence before an agent claims code is duplicated, unused or disconnected.
CLI + MCP server
- 02Agent integrations
MCP servers that connect agents to real work
Tools for reading and changing structured data. Our scheduling MCP includes conflict detection, backfill and forecasting.
Scheduling MCP source
- 03Review systems
Evidence before approval
Template-review workflows that collect evidence for a reviewer. The published field report shows what worked, what failed and why judgment stayed with a person.
Published evaluation
- 04Marketplace frontends
An interface for the people behind the assets
The Webflow Asset Dashboard brings asset management, editing, analytics and submission status into a creator-facing interface.
Dashboard implementation
- 05Dynamic forms
Catch submission problems at the source
A two-step marketplace submission form with published-URL checks, image uploads and validation feedback before work reaches a reviewer.
Form + validation implementation
- 06Agent skills
Methods another operator can use
Packaged architecture and debugging skills give agents a repeatable way to classify a system, trace a failure and design an integration.
Three-Tier Framework skills
- 07Webflow apps & components
Checks and interfaces inside the tools people use
A Webflow Designer validation app, plus reusable React components for forms, navigation, approval gates and workflow interfaces.
Designer extension + component source
What we’ve built / What we can build with you
Different work.
The same care.
From the agent’s tools to the customer’s screen. Find the work you need, then see how it happens.
Choose a film to take a closer look.
Two recorded client-site walkthroughs, a PCN product overview, three illustrative storyboards, and Ground’s actual MCP result on synthetic files.
Selected work
Work you can visit.
Websites, creator tools and learning experiences. Explore the work and the tasks each one supports.
Cato Insights Hub
An Insights Hub built with Webflow components, bringing research, reports and newsroom content together.
Apps & dashboards
Webflow Asset Dashboard
A workspace where creators can track review status, explore Marketplace insights, and edit their assets.
Recorded with a test portfolio. The editor is opened; no changes are saved.
Education / PCN
A home for the practice.
Private networks for sharing agentic engineering techniques, tools and learning resources.
Outerfields / Prototype & platform development
From prototype to a funded platform.
We built the prototype that helped Outerfields secure funding for its first private content network, enabling a $10,000 development budget. Their team later brought development in-house. The lessons from that work informed our own PCN.
- 01 / PrototypeMake the idea tangible.
Build the first PCN prototype.
- 02 / Funded development$10,000 to develop it.
The prototype helped secure funding for the first network.
- 03 / In-house developmentTheir team takes it forward.
Outerfields later moved development in-house.
- 04 / What we learnedA foundation for our PCN.
Apply those lessons to CREATE SOMETHING’s own platform.
Our role: the prototype and early development. The current Outerfields site may include later work by its in-house developer.
Engineering contributions
We improve the tools we work with.
Independent contributions to CTX and OpenAI Codex Security, with accepted changes and release evidence. Our Webflow review-system report documents the work and its limits.
Selected projects demonstrate our capabilities. Membership delivery is agreed separately, one workstream at a time.
AI-native technical support membership
Choose the support your work needs.
Bring a workflow, software problem or existing product. We diagnose the problem, agree on the change, engineer and test it, then help your team use the result. We retain project context across agreed work.
One workstream
Focused
$900/month
For one agreed workstream at a time, with diagnosis, engineering, learning and operator support.
- 2 × 30-minute check-ins each billing month
- 2 × 60-minute attended remote work sessions
- 1 agreed milestone with 1 feedback and revision round
- Skills, learning resources and Slack support
Up to two workstreams
Team
$2,500/month
For up to two agreed workstreams and more time working with your team.
- 2 × 30-minute check-ins each billing month
- 4 × 60-minute attended remote work sessions
- 2 agreed milestones across up to two workstreams, each with 1 feedback and revision round
- Skills, learning resources and Slack support
A milestone is one agreed workflow map or one scoped improvement to an existing workflow. It includes an acceptance check, verification and a handoff you keep. We agree on scope, session times, response times and any usage budget before payment. Cancel before renewal. Larger Builds, managed Control and production incident response need separate agreements.
You keep the agreed code, tests and instructions under your agreement, with reusable tools licensed for continued use. See how delivery works → What you keep →
Delivery and usage are separate
Know the cost before the work starts.
- Included in membership
- Agreed delivery work, guidance and support. Our everyday coding tools are our expense.
- Budgeted separately
- Project-specific AI development, testing and production usage, hosting and third-party services. You approve the budget first.
- Quoted before we expand
- Work beyond the agreed scope. More API credit does not add unlimited delivery capacity.
For substantial AI work, we arrange client-owned API access or agreed metered billing before starting. No API account is needed for work that does not use AI.
Before we begin
A few practical questions.
Agree on the work, keep what we build, and control the costs.
What does ongoing support mean?+
We keep project context across agreed work: the diagnosis, decisions, changes, checks and open questions. Scheduled sessions and Slack support follow your plan. Scope and response times are agreed before payment; production incident response requires a separate managed Control agreement.
Where do we start?+
Bring one task and an example of a good result. We agree on the deliverable, capacity, revisions and how to check the work before payment. If the task is unclear, start with a mapping session.
How do you check the work?+
We test an agreed example and a failure case. For AI agents—software that uses AI and connected tools to carry out a task—we agree on what AI may do and what needs approval. Your team reviews the result before launch. Launch and production incident response require a separate agreement.
What do we keep?+
After full payment, you own commissioned deliverables under your agreement, except reusable provider tools. You keep the instructions, tests, project history and recovery guide. We help your team use and maintain what we deliver. See Services for the delivery process and What You Keep for ownership and provider details.
How is AI spending controlled?+
For substantial AI development, testing or production, we arrange a client-owned provider project or agreed metered billing. Development and production have separate budgets. We pause new billable runs at the agreed limit until you approve more. A website or form that does not use AI needs no API account.
Can we cancel?+
Cancel before your next renewal by emailing [email protected] to stop future membership charges. Support continues through your paid period while we hand over your code and documentation, document open work, and plan the transition. If you need more help afterward, we’ll agree on the scope and price before extending support. Separately scoped projects follow their agreed terms.