FAQ

Yakumo is a development and consulting practice built around AI agents. We are a small team running contracted development and our own products in parallel — our day-to-day operations are driven by a network of agents with distinct responsibilities, all built on Claude Code.

What we offer is not strategy-on-paper, but the agent design patterns and operational knowledge we run inside our own organization. We do not stop at consulting or at development alone — we see it through from concept to working implementation.

There are two points of difference. First, unlike firms that stop at strategy consulting, we speak through implementation — not just ideas. Second, unlike self-described "AI companies" that merely call OpenAI APIs, we actually operate an AI-centric organization ourselves.

The After state our clients are working toward — an organization where agents with distinct responsibilities drive daily operations — is precisely the current state Yakumo already operates in. That is why we can offer a reproducible path, not a theoretical one.

Yakumo is run by a small team: one PM who also works as a junior engineer, and one senior engineer. The PM writes code and designs agents alongside the senior engineer. We deliberately keep implementation and client-facing work on the same hands, so design judgment stays connected to the realities of the work. For full details, see About Yakumo.

We run three internal products: Cortex (a cross-functional agent organization), montage (a video content generation pipeline), and laplace (a financial data substrate, formerly medallion). All three are designed, built, and operated entirely by Yakumo — agent definitions, skill sets, integration scripts, and configuration files are all written in-house.

The underlying LLM is Claude, accessed via Claude Code. We do not present this as "owning a proprietary LLM." The value lies in how we have designed agent organizations and business systems on top of Claude Code — and that design knowledge is what carries over into client work.

These products are internal infrastructure, not commercial offerings.

Claude Code offers flexible agent and skill definitions, direct access to frontier models, and a configuration-driven structure that lets us accumulate knowledge as a durable asset. Reproducing that same flexibility on a different stack would not be worth the cost.

We do implement some engagements on other stacks, but the design judgment behind those decisions is grounded in what we have built on Claude Code.

Our work falls into eight service areas: AI agent development, AI-native system development, AI bot development (chat and voice), AI strategy consulting (from executive decision-making through organizational adoption), business system development, mobile app development, workflow automation (GAS, Chrome extensions, RPA), and website & LP development.

All of these are positioned within an AI-driven workflow design context. Even conventional work like web production or GAS automation is taken on as a component of a larger shift toward AI-centered operations. See the Services overview for details.

Yes, there are clear categories we decline: "AI-flavored" cosmetic changes to existing SaaS products, automation that poses ethical problems, large-headcount fixed-price implementation work, and system replacements that assume no AI involvement.

We place our value in redesigning operations with AI at the center — so requests that do not align with that design philosophy are not a fit. That said, supporting clients who do not yet know where to start is one of our strengths, so reaching out before requirements are defined is fine.

The criterion is procedural determinism. Work that can be expressed as a deterministic procedure goes into scripts or GAS; work that involves probabilistic reasoning, natural language, or branching judgment goes to an LLM; and work that requires final approval or ethical judgment stays with a human. We draw the line across those three zones.

Handing everything to an LLM leaves variance. Yakumo's default flow is "AI produces the first draft, human confirms."

Yes. We support this through a three-part package: Claude Code onboarding, AI literacy training, and ongoing advisory support. Delivery is not the end — leaving operational capability on the client side is part of the job.

Agent definitions go stale within months, so building the people who can keep improving them internally is what we prioritize.

Yes. We use three approaches depending on the situation: inserting LLM inference into existing business systems via API, placing a bot in Slack or Teams, or layering AI functionality on top of an existing SaaS product via Chrome extension.

Even when a full replacement is not realistic, there is usually a path to embedding AI into operations while preserving existing assets.

Our track record includes workflow automation and digital transformation work in manufacturing, AI agent and marketing agent development in the reuse/recommerce industry, and sales agent development for a mobile carrier.

Our strength is less about industry specialization and more about the design pattern of placing AI at the center of operations — a pattern that transfers across sectors. For new industries, we typically start with a PoC-scope engagement to validate the approach before scaling.

We offer eight services. In the AI domain: AI agent development, AI-native system development, AI bot development, and AI strategy consulting (covering executive decision-making through organizational adoption). In the web domain: corporate site development and LP production. In the business systems domain: application development and workflow automation.

All eight are accessible from the Services hub. While the AI services and the web/systems services may look unrelated, we take them all on within the same context — as components of AI-driven workflow design.

It depends on where you are. If you need to start at the executive level — AI investment decisions, organizational adoption, and the full design — AI Strategy Consulting covers strategy, Claude Code onboarding, and training end-to-end. If you want to start by improving a specific workflow, Workflow Automation (GAS / Chrome Extension / RPA) is the right fit. If you want to stand up a conversational AI interface, AI bot development is the entry point.

If none of those fit clearly, send us a message through the contact form. A 30–60 minute intake conversation is usually enough to identify the right starting point.

It means designing and building a network of agents that complete tasks autonomously. This goes beyond conversational interfaces — it includes ambient agents that run continuously in the background of your operations.

The design starting point is the Cortex pattern Yakumo runs internally: a structure of accountable agents combined with a set of deterministic skills that together drive ongoing work. See AI Agent Development for details.

The difference is that AI inference sits at the architectural core, not as a feature bolted on. In addition to human-facing UIs, we design CLI and API interfaces from the start so that external AI agents can also operate and consume the system.

This applies to situations where retrofitting AI onto a conventional system would create friction — for example, systems where judgment branching or natural language input runs continuously. See AI-Native System Development for details.

This is offered as part of AI Strategy Consulting. It covers agent definition design, skill set development, team rollout planning, and operational rule-setting. Deliverables are a reproducible configuration set (the contents of `.claude/agents/` and `.claude/skills/`) and an operations manual.

Combined with strategy work and training, the goal is to leave a state where the client can keep improving independently. See AI Strategy Consulting for details.

Two points stand out. First, delivery speed: with Claude Code at the center of implementation, we move from start to delivery in as little as one week, and rarely more than a month for larger scope. Second, the design assumes AI agents will participate in ongoing operations from the start — content updates and article creation are structured to be handled by agents.

Corporate sites, LPs, and applications are positioned as components of AI-driven workflow design. See Website & LP Development and Business System Development for details.

Yes. Automation that does not put an LLM at the center is still taken on as groundwork for AI-driven workflow design. There are cases where building impact with GAS first — before introducing AI — is the realistic roadmap, so we approach this in stages.

We have track record in manufacturing workflow automation. See Workflow Automation (GAS / Chrome Extension / RPA) for details.

The standard flow is five steps.

1. Contact — Send your inquiry through the form
2. Discovery — One or two 30–60 minute sessions to understand your current state and requirements
3. Proposal and estimate — We submit a proposal covering scope, timeline, cost, and contract structure
4. Contract — We execute an NDA, a master services agreement, and a project-specific statement of work
5. Execution and delivery — We run the engagement through regular check-ins and async communication, delivering the agreed outputs

For larger scope, we recommend phased contracting: PoC scope first, then MVP, then production.

Typically one to two weeks, covering one or two intake sessions (30–60 minutes each), internal review, and proposal preparation.

For smaller engagements with clear requirements, it can be faster. For engagements that need more requirements work, the discovery phase alone may take two or more weeks.

It varies considerably by scope, so there is no single answer. As a rough guide: PoC scope is typically 1–2 months; web site projects start at around one week; and bringing an MVP to production-ready quality generally takes around three months.

For agent development and AI-native systems, we recommend phased contracting — validate feasibility and impact through a PoC first, then move into MVP.

The default structure is weekly check-ins (30–60 minutes), continuous async communication via Slack or Discord, and GitHub PR reviews. We run text-first and async by default, concentrating synchronous sessions at decision points.

This can be adjusted to fit the client's existing communication practices.

Yes. Supporting clients who do not yet know where to start is actually the core territory of AI Strategy Consulting.

A common pattern is to start an AI Strategy Consulting engagement before requirements are set, and spend three months building the requirements and implementation plan together.

Yes. We support this through a hybrid contract (fixed-price initial delivery plus a monthly retainer) or a time-and-materials monthly engagement.

Agent definitions go stale over time, so we recommend structuring for continuous improvement from the start. Dropping Yakumo's involvement to zero immediately after delivery is not realistic for agent-based work.

We use four contract structures depending on the engagement.

• Fixed-price contract: A fixed amount per deliverable. Used for site production, LP, script development, and other engagements with a well-defined scope
• Time-and-materials contract: Monthly or hourly. Used for consulting, advisory support, and ongoing operations
• Hybrid contract: Fixed-price initial delivery plus a monthly retainer. Common for web projects and engagements that include ongoing operations
• Ongoing engagement (employee-equivalent commitment): For long-term, deeply embedded work

Payment terms: fixed-price engagements require a 30% deposit at kickoff with the balance due on delivery; time-and-materials engagements are billed end-of-month with payment due the following month.

Starting at $10,000+ for PoC scope and $30,000+ for MVP scope (production-ready quality for external deployment or high-volume operations).

Final pricing depends on the agent's scope of responsibility, the number of external systems involved, and operational volume. We recommend phased contracting — validate feasibility at PoC scope before moving to MVP.

Starting at $10,000+ for PoC scope and $30,000+ for MVP scope.

Final pricing depends on system size, external integrations, and operational requirements. Designing an architecture with AI inference at the core differs significantly in effort from conventional system development — we recommend a PoC-first phased contract to lock in the design direction before committing to full scope.

AI Strategy Consulting covers strategy formulation, Claude Code onboarding, and AI training end-to-end. Price ranges by scope are as follows.

• Strategy formulation (consulting): $15,000–$50,000 for a 3-month package, or $5,000–$15,000 per month on a time-and-materials basis
• Claude Code onboarding: $5,000–$20,000 initial engagement plus $3,000–$10,000 per month for ongoing support
• AI training: $3,000–$10,000 per session

Final pricing depends on organization size and deployment scope. When bundling multiple scopes, we will provide a custom estimate.

Starting at $10,000+ for PoC scope and $30,000+ for MVP scope.

Final pricing depends on the knowledge scope, integrations, and expected traffic volume.

Starting at $10,000+ for PoC scope and $30,000+ for MVP scope.

Final pricing depends on feature scope, data volume, and whether integration with existing systems is required.

• GAS development: $1,500+
• Chrome extension development: $1,500+
• RPA development: $5,000+

Engagements with a clearly defined scope can start quickly and at a lower initial investment. We have track record in manufacturing workflow automation.

• Website: $1,000+
• LP (landing page): $1,500+

Pricing varies considerably based on page count, design requirements, and implementation complexity (CMS integration, multi-language support, AI integration, etc.). We will provide an estimate after an intake call to understand your specific requirements.

• Fixed-price contract: 30% deposit at kickoff, remaining 70% on delivery
• Time-and-materials contract: End-of-month billing, payment due the following month

If your procurement process requires a different payment schedule, please raise it at the contract stage.

Yes. We can sign before the initial inquiry or before discovery sessions — whichever timing you need. We can work from your template or provide ours.

The industry examples we reference in our track record are all governed by NDAs; individual client names and proprietary details are protected accordingly.