Introducing Diaflow - an autonomous AI teammate platform

Hello Tinhte brothers, this is my first post after 17 years of being a member. In the past, I joined Tinhte when I was in 8th grade, in 2008, to tinker with some BlackBerry Rom mods. After that, I first met Uncle Hiep (I used to call him Hiep = uncle 😁) who borrowed the Mini-DVI to VGA output cable to give a presentation for the Youth Informatics contest in 2009 at Cafe Tinhte when I was still at Ban Co, at that time I was studying at Colette school. In the blink of an eye, nearly 2 decades have passed, since the time when we were still diligently modding BlackBerry and Symbian ROMs, tinkering little by little. Today, I'm not back to ask questions about phones, but to write this article to introduce Diaflow, an AI platform that my team and I have devoted our efforts to building over the past year. Our starting point is a node-based startup working on Workflow Automation. However, after a period of rebuilding and rebuilding, we are now switching to developing AI agents to help users solve tasks much easier than sitting around "laying bricks" to build a workflow in the old style. Below is our story and why Diaflow was born. This is our team in Vietnam (photo taken in November 2025 while on a company trip in Da Lat).

1. The pain of Workflow Automation and the turning point of AI Agents

In fact, with the Workflow Automation segment, Diaflow has helped many business customers build extremely stable automation systems. We handle office workflows such as automatically extracting invoice and document data, to highly creative workflows such as automatically producing movies, making advertising TVCs, or even making music. The example below is a simple workflow to create a small scene in a video.

If you've ever used automation flow tools (like Zapier, Make or n8n), you'll find it very powerful. But the harsh truth is: not everyone can do it. It's quite confusing for those who don't have programmer-like logical thinking. Every time we want to automate a process, what do we do? Having to connect multiple nodes, map data from one field to another, struggle with API keys, and roundabout error handling. Looking at a screen full of wires is sometimes more stressful than doing it manually. Last year, when large language models (LLMs) with inference capabilities began to explode, our team realized a truth: Why make users manually build each step of the workflow, when we can create "AI assistants" (AI Agents) capable of understanding requests in natural language and automatically reasoning and executing those tasks? Instead of having to drag and drop 10 nodes to read mail, filter information and fill in an Excel file, now you just need to tell the AI: "Read this pile of mail, filter out customer information and export me a .xlsx file". Currently, Diaflow Intelligence can effectively handle many different tasks, from accessing and synthesizing information across applications and workflows, to creating complete products such as spreadsheets, presentation slides (can be edited right in the browser instead of having to download files to edit), documents or web applications. In addition, the system also has the ability to automatically break down large projects into steps and execute them sequentially. Here is an example of a task that Diaflow performs: use /deep-research to research a topic and create slides for a presentation. Here everyone can see that Diaflow Intelligence will automatically read the skill /slide-generation to choose the style that best suits this topic (these styles will be integrated in this skill folder) as well as scripts to create to-dos, assembly files, etc. I'm proud to rate the visual quality better than Claude hehe.

2. What's different about Diaflow Teammates? ("Virtual HR" ecosystem)

My company spends a lot of money to use AI and recruit people, partly because it has many credits sponsored by companies like AWS, Azure, GCP because it is one of the portfolios of Insignia Ventures Partners fund (venture capital fund at Tier 1 level in Southeast Asia); The rest is to scale the company and revenue. However, after a period of use, I realized that agents always have to wait for the team to interact before working or recruit too many and there is a bottleneck in human resources: experienced human resources are expensive, too young, they are difficult to manage. This is when our team decided to build Diaflow Teammates - a platform shaped according to the AWaaS (AI Workforce as a Service) model. Surely you have heard of the popular keyword One-person company recently, which is "1-member organization".

3. So how is Diaflow different from using ChatGPT or Claude directly?

Surely many of you will ask: "So I use ChatGPT or Claude directly, why do I need Diaflow?". The core difference lies in the independent Multi-Agent Architecture. When you use ChatGPT, you cram everything into one chat. Today ask for code, tomorrow start writing sales emails, day after tomorrow start doing marketing. As a result, the AI's memory (context) is bloated and disturbed. It begins to get "overwhelmed" with information (out of context), remembering, forgetting, and worst of all, hallucinations (hallucination - lying, making up parameters). Not to mention, tokens are burned horribly because the agent has to reason (think). With Diaflow, we solve this problem by:

4. Hidden Power: Agent Runtime, Execution Sandbox and Routines

In order for these Agents to not just "talk" but actually "do the job", we equip the system with an extremely powerful architecture including Agent Runtime, Sandbox and Routines (what people are calling Harnessing these days):

Credit: Teamcopilot

5. Why does Diaflow target the Global market instead of Vietnam?

Based on the actual data of the platform, we realized that 85% of customers are using Diaflow for repetitive tasks: extracting data from files (OCR/Vision applications), transferring data between platforms, data entry... In the US market, these jobs are often assigned to juniors or freshers with an average salary of about $60,000/year. When using Diaflow, businesses can replace the labor of more than one person and be much more efficient. The ROI problem (Return on Investment) when applying AI in the US is extremely clear. On the contrary, in Vietnam, $60,000 (about 1.5 billion VND) is a very good salary fund to support a junior team of 10 people for a year. Therefore, using AI to replace human resources in Vietnam currently does not bring economic efficiency that is attractive enough for businesses. That is why our team is currently focusing 100% on the US market and English-speaking countries (English-first).

6. Diaflow's vision: Lean more, go further

Many people are afraid that AI will take away their jobs, but at Diaflow, we do not believe that AI can completely replace humans. Instead, AI was born to help people achieve ambitions that previously required a lot of resources, money and time to realize. In the future, I think a company's structure will be much "leaner" thanks to AI. There will no longer be cumbersome machines filled with personnel handling data manually. Instead, 1 person with the help of AI can create the same value as 10 people before, and of course, that person will receive more benefits and have time for more strategic and creative decisions.

7. Gifts exclusively for Tinhte brothers

Even though I went to the ocean, my roots were still Vietnamese, I was still the kid who used to walk from Colette school to Cafe Tinhte in Ban Co to wait for my seniors. The products I make, I always want to bring back and share with the community first to listen to sincere opinions (and even frank criticism) from people who love technology. If any of you are feeling overwhelmed with repetitive work, want to find ways to automate but are lazy to learn code, or simply want to learn how a real "AI company" works, you can stop by and try out our team's products. Product link: https://diaflow.io/ As a thank you, I would like to send you 100 discount codes: `TINHTE26` corresponding to $192 for subscription packages or credit purchases. If you have any comments, questions, or comments, please enthusiastically comment below. I will directly read and record everything so the team can improve the product better and better! Thank you for taking the time to read this thoughtful article. Anyone interested in connecting with me can connect with me via Facebook or Linkedin: