Day 32 of 100 Days Agentic Engineer Challenge: Perfect Flow and Smart Bridge

I wonder, are we ready for fully autonomous agentic systems without our interference? Do we need to build a new AI agent or will it be a part of new features of existing platforms and extensions. Why build a social media marketing agent when platforms like HubSpot can extend their services and offer AI agent as a service. With the amount of data they have, no one else will be able to build better agent, at least for inbound marketing tasks. I have a lot of questions to answer, but let’s start with my daily tasks.

Daily Tasks Routine

1. Physical activity — I did 35 push-ups again, so I managed the whole January 2025, every day push-up exercises.
2. Seven hours of sleep — I slept for 7 hours, but went to sleep too late.
3. AI Agent — I decided to finish my AI image and video generation platform, which I started as an educational project while learning a low-code approach to building apps. It has nothing to do with an AI Agent yet, but I am thinking about a semi-automated flow with a human in the loop.
4. PAIC — In queue.
5. Data Science — In queue.

If you want to know what all these tasks are about, read the introduction to the 100 Days Agentic Engineer Challenge.

Are we ready for fully autonomous AI Agents?

I think not yet, maybe if we can do it with a semi-automated approach with approval, it can work. But it’s interesting that everyone is talking about AI agents and a lot of SaaS platforms don’t even offer simple task automation. I come to a social media scheduling platform where I have to connect each social media account, upload my media, and even create each post and schedule it manually. There are some platforms where it’s enough to add a link to business website and the platform make posts preparation for us, but it doesn’t work very well yet. I think first the existing platforms need to adopt automation workflows with good results before we can start building AI agents.

Diffusion Models Aggragator

Let’s take another example: AI image and video generation and a platform called Freepik. It’s an aggregator of different distribution models for images and videos. We login, add a prompt and generate an image, we can also generate a video based on an image that we can upload or choose from already generated. The last feature is nice, we have a small connection between image and video generation interface, but what if we have hundreds of generated images in different styles from different days or months? Probably we need to manage the organization of the media locally and get some consistency in image gen using ChatGPT to generate some prompt sequences, but still aggregator is a good thing because we have all the different diffusion video models in one place, so the number of required fields is reduced. What could we improve or from a founder’s perspective, should we build an agent that reduces the number of tools and decisions even more and try to compete with pretty big players on the market?

Resistance to Competition

If we decide to make an image and video gen SaaS, that solves similar problems as other platforms available. How can we compete with them? The best way is to niche down, the image and video gen only for fashion industry or for real estate, but how to compate with other much bigger portals without narrowing the target groups? There is actually one thing we talk about all the time and the solution is to reduce the number of activities the user has to take to get the same result as from competition, it’s maybe not an AI agent, but is going in a similar direction, what is the purpose of AI agents? To take over as much work as possible from the human.

Pixonaut.art — Perfect Flow and Smart Bridge

As an educational project I built in 4 weeks an image and video generation platform with possibility to use different diffusion image and video models from one interface. But then I thought, now I’m nothing else than a next aggregator platform. Then the idea was to go more into cinema and commercials, but still you can use Freepik or MidJourney with RunwayLM to manage any kind of content. The final solution was of course based on reducing the amount of user activity and integration and making a perfect flow with connecting all the different media interfaces together. I created a story generator with all the prompts needed to generate all the media (image, audio, video) for an AI narrative movie or documentary with the possibility to pass it to different stages of final video production (smart bridge). It still needs a human supervision, but reduces the user interaction to writing a small description for the whole movie or commercial.

I wanted to build an AI agent, but why not improve my existing applications with automated or semi-automated workflows, and then after that works, slowly move to agentic systems.