During my time working with Jutsu, I gained valuable insights into the complexities of building AI tools that integrate seamlessly into the day-to-day operations of businesses. What stood out to me about Jutsu was their mission to make AI applications reliable, secure, and user-friendlyâaddressing a significant gap in the AI space as more businesses adopt these tools. I worked on two key features: the knowledge base and the prompt studio, both of which were designed to make AI applications more grounded in real-world business needs.
We need âPerplexityâ for private data because so much data is hidden. Whoever gets the plugins and indexing right will take off because the existing search on platforms sucks.
One of the main takeaways from my time at Jutsu was the importance of having a âsingle source of truthâ within a business. Many organizations work with siloed dataâsales, customer service, marketingâoften disconnected and difficult to manage. Jutsuâs knowledge base sought to solve this problem by consolidating all internal and external data into one unified system. This didnât just mean collecting data from different sources; it meant making that data actionable, so it could directly inform AI-driven actions. The power of the knowledge base, as I learned, was its ability to process and activate data in a way that was genuinely useful for the business.
The prompt studio, which I was also involved with, helped optimize the reasoning pathways that AI uses to perform business tasks. Instead of simply giving an AI a task, you break it down into smaller, manageable actions that guide the AI step-by-step. The prompt studio allowed us to chain these tasks together in a way that ensured each action was aligned with the business's goals. It wasnât just about making sure the AI did what it was told; it was about making sure it did the right thing in a way that was aligned with how the business needed to operate. This approach helped the AI feel more like a useful tool than a black box.
Letâs understand how generation works.
1. LLM - Choosing the right Large Language Model to power your application.
â
It is becoming less of a headache the further we are into the LLM craze. The performance of available LLMs are converging, both open source and proprietary. The main choice nowadays is around using a proprietary model or self-hosting.
2. Prompt Engineering - having context available for usage in your prompts does not free you from the hard work of engineering the prompts. You will still need to align the system to produce outputs that you desire and prevent jailbreak scenarios by Guard-railing inputs and outputs.
What did Jutsu fit in?
Context is all you need.
The hard things for humans are easy for machines. The things we take for granted are insanely hard for machines. This is not a new idea. It is called moravec's paradox in robotics and it is now the main reason why companies are not seeing enough ROI on AI.
AI excels at complex, abstract tasks like chess, but struggles with simple everyday tasks like vacuume cleaning your house.
Cleaning your house, folding your laundry.. None of these things require a PhD, but they require:
- Perception
- Context awareness
- Real-time physical interaction
- Common sense
This is the AI context paradox. Putting things in the right context, building on intuition we built over many years to make sense. This is the key observation for differentiated value. You need deep interprise expertise context.
General assistants are for convenience. Predefined enterprise workflows are for efficiency. Specialized agents.. now that is unlocking ROI.
Right now, everyone is focused on convenience and maybe some decent agentic workflows. This is great but let's be real, we are not at agent level yet. Don't let the hype fool you.
How do we aim at differentiated value?
LLMs are awesome but they are 20% of a big system. Everybody heard of RAG. Everyone is thinking about the LLM but very few people think of the system around it. That system is what solves their problem.
We have to think of systems. Not models. If we want ROI. In an enterprise, expertise is your fuel. Date is moat.
Pilots are very easy. The gap between pilot and production is always larger than you expect. Sure you can show your boss an MVP of a customer support chatbot but once you are asked to productionize it.. now you need to think of edge cases, thousands of users, security risk, enterprise requirements for compliance.. This is thinking at scale.
How to approach this? Don't design for pilot. Design for production. But at the same time, speed is more important than perfection. So give it to your users relatively early to get their feedback. And at the same time we have engineers working on really boring stuff like what is the ultimate chunking strategy for my RAG system and how is it different for every use case... At the same time everyone wants engineers to think about differentiated value.
There is business priorities and there most time consuming tasks. Consolidating both is insanely hard.
And the craziest part is that after all of this.. very often people do not use the AI that their company built for them and bring their own outside AI tools to the workplace even though their company built company-native AI integrated in their flow. They still prefer to use ChatGPT for example feeding company data to it.
Â
How about wrappers?
What distinguishes successful wrappers from their less successful counterparts: intuitive UI, Propriatary evals, finetuning of foundational models and thoughtfully designed multi agent architectures.
Â
Working at Jutsu shaped my thinking about AIâs role in business. One of the key lessons I took away was that AI, in its current form, is more effective as a tool to enhance human efforts than as a complete replacement. Businesses donât need AI to sell for themâthey need AI to help them sell smarter. The AI agents we built were designed to assist with tasks like customer acquisition, marketing campaign execution, and task management, all with the goal of helping the business owner leverage their own vision and expertise. This philosophy set Jutsu apart from other AI solutions, which might focus on automating everything for the user. Instead, Jutsu aimed to enhance the business ownerâs ability to make decisions and manage operations.
The core value of AI, in my view, isnât just in its outputs but in its ability to embed itself into existing workflows and improve over time. At Jutsu, we werenât just selling an AI agent; we were offering a way for businesses to integrate AI into their daily operations. The long-term goal was to create a platform that could securely handle vast amounts of data, train AI systems using that data, and continuously improve the agents over time. This would not only make businesses more efficient but would also position Jutsu to deliver even more powerful AI tools in the future.
One of the things that became clear to me at Jutsu was that AI should be designed with the user in mind. Thatâs why we focused on making the tools intuitive and easy to use, ensuring that businesses of all sizes could adopt them. But Jutsu didnât just aim to make AI accessibleâit was about building systems that would allow businesses to manage and leverage their data better. By working with a variety of clients, we could gather the necessary insights to refine our tools and make them more effective.
The real value of AI lies in its ability to help businesses manage their data-driven relationships with customers. AI isnât about creating magic behind the scenes; itâs about providing business owners with the tools they need to unlock the full potential of their own data. A business is an extension of its founder, and with Jutsuâs agentic system, those founders can extend their capabilities further, streamlining workflows and increasing efficiency.
Jutsuâs mission to build reliable, secure, and intuitive AI tools for businesses resonated deeply with me. The knowledge base and prompt studio were core features that allowed businesses to integrate AI into their workflows and leverage data to drive smarter decisions. My time at Jutsu reinforced my belief that AI should be a tool to enhance human potential, not replace it, and that the most effective AI tools will be those that can embed themselves into existing business processes and continuously improve as those processes evolve.
Â
Â