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.

GPT-4 Just Dropped — Here's What It Means for Your Business (Not What You Think) — BlokAI Blog

"GPT-4 is OpenAI's most capable and aligned model yet, capable of passing the bar exam in the top 10% of test takers." — OpenAI Blog, March 14, 2023

GPT-4 is out. The benchmarks are stunning — bar exam top 10%, advanced biology, passing coding assessments that would filter out most junior developers. OpenAI's March 14th release dropped a technical report that reads like a quiet declaration that the rules have changed.

Here's what I've been telling clients over the past week: calm down, and then get excited about the right things.

What Actually Improved

Let me be specific, because the hype machine is already cranked to eleven and most of what's being written is either breathless or dismissive.

Reasoning quality jumped significantly. GPT-3.5 (what powers the free ChatGPT) would often produce plausible-sounding nonsense when pushed on complex, multi-step problems. GPT-4 is meaningfully better at staying on track through longer chains of logic. For business use cases that involve analysis, synthesis, or structured decision-making, this matters. Context window expanded. The 32k token version of GPT-4 can hold approximately 25,000 words in a single conversation. That's a long document, a whole meeting transcript, or a substantial chunk of your product documentation. This changes what's possible for document-heavy workflows. Instruction following improved. Getting GPT-3.5 to reliably follow complex, multi-part instructions required significant prompt engineering effort. GPT-4 is more obedient, which translates to more predictable behaviour in production systems. Multimodality is coming. The technical report describes image input capabilities that weren't released publicly — yet. This is a significant signal about where things are heading.

What Didn't Change (That You Should Know)

It still hallucinates. GPT-4 makes things up with confidence, just less often than its predecessor. Any production system you build still needs validation, guardrails, and human review on high-stakes outputs. Don't let the better benchmarks create overconfidence. It's still a stateless API call. GPT-4 doesn't remember your business, your customers, or your previous conversations (unless you explicitly feed that context back in). The memory and personalisation problem hasn't been solved by a better model — it requires better architecture. Cost went up. At launch, GPT-4 via the API costs roughly 15x more than GPT-3.5-turbo per token. For high-volume applications, this is a real consideration. Not a blocker, but it changes the economics.

The Practical Business Implication

The gap between GPT-3.5 and GPT-4 is large enough that it reopens some use cases that were marginal before.

Complex document analysis — legal contract review, financial report summarisation, technical specification parsing — becomes more viable. The accuracy bar is higher, and the failure modes are less frequent. You can build with more confidence.

Customer-facing applications become somewhat safer to deploy. Better instruction following means the model is harder to jailbreak and more likely to stay on topic. Still needs guardrails. Still needs monitoring. But the baseline is better.

Code generation quality improved meaningfully. If you're building or evaluating internal tools, AI-assisted development is now a serious productivity multiplier rather than a novelty.

Who Should Upgrade, and When

If you're currently using GPT-3.5 in a production system and it's working well enough: don't rush. The cost increase is real, and "working well enough" is often the right bar.

If you've tried to build with GPT-3.5 and hit walls — the reasoning wasn't reliable enough, the instruction following was inconsistent — evaluate GPT-4. For the right use cases, the jump is significant enough to unblock you.

If you haven't started yet: GPT-4 is the baseline I'd recommend building against for anything customer-facing or analytically complex. Start with the assumption that you'll pay for quality, and build your unit economics accordingly.

The Bigger Picture

We're two major releases into the post-ChatGPT era, and the pace isn't slowing. GPT-3.5 launched in November 2022. GPT-4 in March 2023. Both Microsoft and Google are racing to embed these capabilities into their existing product lines. Open-source alternatives are advancing faster than most expected.

The businesses that will be well-positioned in two years aren't necessarily the ones with the most sophisticated AI — they're the ones that have built the internal capabilities to evaluate, adopt, and integrate AI tools as they evolve. That means having someone who understands the technology, having your data in order, and having established patterns for how AI fits into your workflows.

That's less exciting than a benchmark score. But it's what actually compounds.

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