How much does AI consulting cost?

BlokAI's engagements range from $5,000 to $75,000 depending on scope. Simple automations and LLM integrations start around $5K–$15K. Custom AI agent systems and full data infrastructure projects run $30K–$75K. We scope every project before committing to a price.

What does BlokAI build?

BlokAI builds AI agents, LLM applications, RAG systems, automation pipelines, and data infrastructure for startups and SMBs. We take projects from architecture to production — not demos, not prototypes.

Is BlokAI based in Japan?

Yes. BlokAI is a solo studio run by Kent Tran, based in Tokyo, Japan. We work with clients globally in English.

What is RAG and why does my business need it?

RAG (Retrieval-Augmented Generation) lets AI systems search your own documents, databases, or knowledge bases before answering — instead of relying only on training data. It powers AI customer support, internal search tools, and sales enablement systems that are accurate and up-to-date.

How AI Agents Are Different From Chatbots (And Why It Matters)

"Sora can generate videos up to a minute long while maintaining visual quality and adherence to the user's prompt — a significant leap in video generation capability." — OpenAI Blog, February 15, 2024

Every week I have a version of this conversation: a business owner asks about "AI agents" after reading something about them, I ask what they mean by agent, and the answer is usually something that's actually a chatbot with extra steps.

The distinction matters. A lot. Getting it wrong leads to expecting the wrong things from a system, building the wrong architecture, or spending money solving a problem that didn't need the solution you built.

Let me clear this up once and for all.

Chatbots: The Request-Response Model

A traditional chatbot — even a sophisticated one powered by a large language model — follows a simple request-response pattern:

User sends a message
System processes the message (looking up information, running through a model, formatting a response)
System returns a response
Interaction ends (or loops back to step 1 for the next message)

The system is reactive. It responds to inputs. It doesn't initiate actions, remember context across sessions (without explicit design for memory), or work toward goals on its own.

Even the best LLM-powered chatbots — the kind that can answer complex questions, cite sources, maintain conversational context — are fundamentally reactive systems. They wait for input, they respond, they wait.

This is genuinely useful. An internal knowledge base assistant that answers employee questions is a chatbot. A customer support tool that handles common inquiries is a chatbot. For many business use cases, a well-built chatbot is exactly the right tool.

Agents: The Goal-Directed Model

An AI agent is different in a fundamental way: it operates toward goals, not just responses.

An agent can:

Plan a sequence of actions to achieve an objective
Use tools — search the web, query databases, send emails, make API calls, run code
Make decisions about which action to take next based on intermediate results
Work autonomously through a workflow without human input at each step
Adapt when something doesn't work as expected

Where a chatbot answers the question "What's in our inventory?", an agent might: check the inventory system, notice certain items are low, look up supplier lead times, generate a purchase order, and file it — all in response to a single instruction like "Make sure we don't run out of product X."

The key property: agency. The system is initiating actions, not just responding to them.

A Concrete Example: The Same Task, Two Approaches

Task: "Get a summary of recent customer feedback and identify the three most common complaints." Chatbot approach:

User asks the question
If connected to the right data, the chatbot retrieves relevant feedback and summarises it
User gets a response
If the user wants more, they ask another question

Agent approach:

User gives the instruction once
Agent accesses the support ticket system, NPS survey results, and recent reviews
Agent synthesises across all three data sources
Agent identifies patterns, groups complaints into categories, ranks by frequency
Agent writes a structured report with specific examples
Agent might also check if there's an existing ticket open for the top issue and add a note
Agent returns the finished artifact, ready to use

The chatbot answers a question. The agent completes a task.

Why This Matters for Building

The architectural difference is significant:

Chatbots:

Relatively straightforward to build and test
Predictable behaviour: input → output
Easier to evaluate (did it give the right answer?)
Lower failure risk (each response is independent)
The right tool for Q&A, retrieval, and conversational interaction

Agents:

Significantly more complex to build reliably
Less predictable: the agent is making a series of decisions
Harder to evaluate (did it achieve the goal correctly?)
Higher failure risk: a bad decision at step 3 can compound through steps 4-8
The right tool for multi-step workflows, task automation, and goal-directed work

The current limitation of agents is reliability. Language models make probabilistic decisions, and in a multi-step workflow, errors compound. A well-designed agent with appropriate guardrails and narrow scope can be highly reliable. A poorly designed agent attempting too broad a goal will fail unpredictably.

This is why narrow, well-scoped agents are shipping in production while fully autonomous general agents are still largely in the research/demo phase.

What's Actually in Production Right Now (Early 2024)

The agent applications we're seeing in production:

Research and monitoring agents: "Monitor these 10 competitors' websites for pricing changes and send a daily digest." Runs on a schedule, uses tools to fetch and compare web content, generates a structured report. Data pipeline agents: "When a new form submission comes in, validate the data, enrich it with third-party data, and create records in our CRM." Multi-step, uses multiple tools, runs automatically on triggers. Content agents: "Given this product update, draft a blog post, summarise it for social, and write the email newsletter version." Still requires human review, but significantly accelerates multi-format content production. Support triage agents: "When a new support ticket comes in, classify the issue, check if there's a known solution, draft a response, and escalate to human if the confidence is low." Handles routine cases autonomously, escalates complex ones.

The pattern: narrow scope, clear tools, defined success criteria, human oversight for edge cases.

The Honest Advice

If you're thinking about AI for your business and someone says "agent," ask what they mean. Are they describing a chatbot? (Often.) A workflow automation with an LLM step? (Also often.) A genuinely agentic system that plans and executes multi-step tasks? (Less often, requires more sophistication.)

Neither chatbots nor agents are inherently better — they're right for different things. Get clear on which one you actually need before you start building.

Want to implement AI in your business? Book a 30-minute architecture call at blok-ai.com.