Skills and agent stacks
Claude Code, OpenClaw, Cursor, Cline, Aider, Codex, Hermes — installed, configured, and matched to the work. Most setups are not one tool, they are two or three that hand off cleanly.
// fractional AI lead
What a month with a fractional AI lead looks like_
The AI release calendar does not pause for your roadmap. Every month there are new models, new harnesses, new integrations — and most operators don't have time to evaluate any of it. A fractional AI lead tracks the firehose for you, audits what your team is using, ships the tools that compound, and trains so it sticks. Part-time senior AI person, on retainer, without the full-time headcount.
What a month looks like
The model is not the bottleneck. Keeping up is.
The models are good. They have been good for a while. What kills most companies on AI is not the model — it is the firehose. New releases every two weeks. Tool upgrades, new harness capabilities, better integrations. Most operators are running a company; they do not have time to read every announcement and decide what matters. So they miss what compounds and adopt what is hyped.
A fractional AI lead is the person who keeps up for you. Month one: audit what your team is using, identify the two or three things that will actually compound, write the roadmap. Month two and on: build the tools that matter, train so they stick, and read every release with your stack in mind. Not a one-off setup — an ongoing relationship with someone who knows your codebase, your team, and your goals.
I do this work hands-on. No course, no playbook PDF, no async Loom dump. I sit with your environment, build the right things, wire the integrations, and stay on retainer so there is always someone who actually keeps up.
What I build and maintain
The five layers that compound over time
MCP servers and CLI integrations
Model Context Protocol servers that plug your internal tools into Claude and friends. CLI wrappers around your existing scripts so the agent can run them safely. The right tools, behind the right guardrails.
Knowledge layer
Your codebase, your docs, your client folders — indexed and reachable from the agent without copy-paste. Retrieval that actually returns the right chunk, not the closest-sounding one.
Role-specific configs for teams
Engineers get one config. Writers get another. Ops and leadership get the views they need. Same underlying stack, different surfaces — so nobody is fighting a tool that was set up for somebody else.
Adoption and workflow standardization
A setup that nobody uses is a setup that does not exist. Real adoption work — short walkthroughs, written norms, a small set of repeatable patterns — so the AI environment survives the first month.
Most retainer months touch three or four of these layers. The audit identifies which ones are gaps. The build work fills them. The retainer keeps them from drifting as the AI landscape shifts.
Service pillars
Three pillars of the engagement
Depending on what your audit surfaces, a retainer month will pull from one or more of these lanes. The duck routes the conversation to the right one.
// AI consulting
→Strategy, roadmap, tool selection, and hands-on AI builds. The engineering arm of the fractional engagement.
// AI automation
→Agents, MCP wiring, and workflow automation that remove the repetitive work from your team's day.
// AI integration
→Claude, GPT, and open-source models wired into your existing stack — APIs, internal tools, the data your team actually uses.
Harnesses I work with
Harness-agnostic. The right answer is usually two of these working together. Each page covers where the tool earns its slot and what configuration looks like.
Claude Code setup
→Anthropic's terminal-native coding agent, configured with the right skills, MCPs, and project context for the way you actually work.
OpenClaw setup
→Open-source agent harness wired into your stack — model choice, tool surface, and run policies tuned to your domain.
Cursor setup
→IDE-first AI coding, configured beyond defaults — rules, indexing, model routing, and the integrations Cursor does not ship with.
Codex setup
→OpenAI's coding agent, plugged into your repo, your environment, and the guardrails that keep it from going off the rails.
Cline setup
→VS Code agent, configured for autonomous loops your team trusts — approval gates, allowlists, and real project memory.
Aider setup
→Terminal pair-programmer, dialed in with the model routing, repo map, and git workflow that match how you ship.
Hermes Agent setup
→Hermes wired into your team workflow with skills, retrieval, and the tool integrations that turn it into a real teammate.
What a month looks like
A fractional month has a rhythm.
These are the recurring pieces of a retainer engagement, not one-off project deliverables. The mix shifts month to month based on what the audit found and what the team needs next.
| Shape | Buyer | What it looks like |
|---|---|---|
| Audit + roadmap | Month one of an engagement | What your team is actually using, what is working, what is noise. A short written roadmap: the two or three things that will compound if we build them this quarter. No 40-page deck. |
| Build the tools that matter | Ongoing retainer work | Agents, MCP wiring, harness configs, integrations with your internal stack. The things the audit identified. Shipped into your repo, your infra, your accounts. |
| Train so it sticks | After each build | Short walkthroughs, written norms, repeatable patterns the team actually follows. A setup nobody uses is a setup that does not exist. Adoption is part of the month. |
| Keep up | What the retainer is for | New models, new harnesses, new integrations — the firehose does not pause. Part of every month is reading the releases, filtering the noise, and telling you what actually changes your stack. |
| Custom scope | Larger or more complex | Multi-team rollouts. Heavy compliance scope. A specific build that needs full-time attention for a sprint. Scoped on the call. |
No public pricing. Retainer scope is set on the call — real monthly rate, real deliverables, real scope. If what you need is a one-off rather than a retainer, the duck says so and quotes that instead.
What I don't take on
Two hard nos. Everything else gets quoted.
Honest list, short on purpose. I would rather you find the right hire on the first try than waste a discovery call on a job I cannot do well.
1. Training AI models from scratch
Not fine-tuning. Not RAG. Actual pre-training of foundation models. That needs a research lab with the compute and the team — somewhere like Together AI, Modal, or a Databricks partner. I would be the wrong hire and you would feel it within a week.
2. Work requiring government clearance
DoD classified work, FedRAMP High needing sponsorship, ITAR, CMMC, anything touching cleared programs. I am a solo Missouri LLC without a sponsor. If you have a contract requiring those certifications, you want a cleared shop, not RDTS.
Everything else — Claude Code installs, MCP wiring, agent harnesses, skills, knowledge-layer integrations, adoption, role configs, custom skills for content teams, retrieval that actually returns the right chunk — get the duck on a call.
I also build software and websites
Retainer is primary. Builds happen inside it.
The fractional engagement often includes shipping real software — Lee De Card, Dragon Wagons, and a longer list on the work page. If your audit surfaces "this thing needs to exist and it does not," building it is part of the retainer. If you need a fixed-price build scoped separately, that is a different shape and the duck will say so.
If you already know you want a full build with a fixed-price scope, the AI developer for hire page covers that lane in detail.
FAQ
Common questions about the fractional engagement
What does a fractional AI lead actually do?
+
Keeps up with AI so you don't have to, then does the work that keeping up reveals. Month one is usually an audit: what your team is using, what is working, what is noise, what two or three things would compound if we built them. Months after that are the build-and-train rhythm: ship the tools the audit identified, show the team how to use them, read the releases with your stack in mind, and show up for the next iteration. Not a one-off setup. An ongoing relationship with someone who knows your codebase, your team, and your goals.
Why retainer rather than a one-off project?
+
Because AI is not a one-off problem. The release calendar does not pause. What is best practice for your stack in January may be outdated by March. A one-off setup gets you to a good place once. A retainer means there is always someone who keeps up — so you do not spend a quarter running on a harness that has been superseded. That said, if you genuinely have a one-off job (a specific build, a broken integration), the duck will quote that shape too and say so plainly.
Are you tied to a specific tool or model?
+
No. Harness-agnostic and model-agnostic on purpose. I default to Anthropic Claude because it earns the slot for most code and content work, and I reach for OpenAI's flagship GPT when it specifically wins. On the harness side, I will set up whatever fits — Claude Code, Cursor, Cline, Codex, Aider, OpenClaw, or Hermes. Most engagements end up using more than one. The opinionated part is matching the right stack to the use case, not selling you whatever I happen to like that month.
What size company is this for?
+
Funded startups and growing companies — typically teams where the operator is too busy to track AI releases themselves but the company is moving fast enough that falling behind matters. The work scales with the team: a three-person engineering team and a twenty-person org need different configs, different adoption work, different retainer scopes. The duck figures out which shape fits on the call.
What does ownership look like?
+
Everything lives in your accounts and your repos. Configs are checked into a repo you own. MCP servers are deployed to your infrastructure. Credentials are yours. Documentation lives in your wiki or your repo, not in a vendor portal. If you want to end the retainer and hand the stack to another engineer, that handoff takes an afternoon — no proprietary DSL, no agency holding the keys.
How is pricing handled?
+
No public numbers. Retainer scope and monthly rate are set on a 30-minute call. You leave with a real number, a real scope, and a real first-month plan — or you leave with an honest "this is not in my lane" and a pointer to who is.
What if I'm not sure whether I need a retainer or a full software build?
+
Most people are not sure on the first call, which is fine. The duck reads what you describe and names the shape. If you are describing something that needs to exist that does not — a custom dashboard, a pipeline, an internal app — that may be a software build rather than a retainer, and RDTS still does those (see /work/ for examples). The shape gets named on the call, not by you in advance.
// Ask the duck
Describe the team. The duck names the shape.
A 30-minute call. You describe what your team is using, what is falling behind, and what you want to be true in three months. The duck reads whether this is a retainer, a one-off build, or something else — quotes the scope, and either books the next step or sends you somewhere better. No scoping form. No qualification grid. No slide deck.