Florian Stegre (@VirtualFlo) header
Florian Stegre (@VirtualFlo) profile picture

Building a bootstrapped AI company from the south of France. I share what it takes to sell to big companies and help them put AI to work.

Bordeaux
Joined January 2010
Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)Photo by Florian Stegre (@VirtualFlo)
I cancelled Figma. Webflow may be next. With ChatGPT and Codex, I now design directly in Checksub鈥檚 code. I also built a new website from scratch. As a bootstrapped founder, this makes me rethink what I鈥檒l keep paying for. What makes a SaaS worth keeping for you?
馃 Made with AI
11
9
205
I also want this setup to make Checksub easier to run and leave me more time for sport. Working with agents takes real energy and focus. I think making time to recover will be essential to keeping up with this new economy. More on that in another post.
We want Checksub clients to build on their existing content when making future videos, with access to different models and the technical complexity handled for them. What would an AI agent need to know about your company that currently lives only in someone鈥檚 head?
1
3
At Checksub, I keep priorities, product context and decisions in a Git repo. My agents update the files as we work, and I load that context into new conversations. The history stays with us when I switch between Claude, ChatGPT or another provider.
1
3
Existing companies have years of knowledge that never gets written down. Why a customer gets an exception. Why a feature works that way. Then we give an AI agent a task and leave that context out. A new employee would struggle with the same brief.
1
20
Talking to several agents, testing their work, and giving feedback day after day takes more focus than I expected. Surfing and wingfoiling are how I switch off. If you build with agents, how much do you ask one to tackle before you test the result?
5
We鈥檝e spent four months rebuilding Checksub with AI agents. Now I give them the kind of brief I鈥檇 give a developer: one feature, a clear scope, and a result I can test. I still give agents room to explore when I鈥檓 trying to figure out what to build.
1
6
I asked an AI agent to add every feature it thought was missing from Checksub鈥檚 video editor. Several days later, I had an unstable editor with a Frankenstein interface. That experiment changed how I assign work to agents.
1
15
I can open the same folder in Claude or ChatGPT, or use it with an open-source model. That lets me pick a model for the job and spread the work across their usage limits. What鈥檚 one piece of context you鈥檇 want every AI tool you use to pick up automatically?
9
I keep a folder of Markdown files in a Git repo synced through GitHub. It covers agent roles, current priorities, how the app is organized, who we build for, and our design system. It also records major decisions and changes so an agent can see how we got here.
1
1
9
I can talk through a decision with my CEO agent in ChatGPT, then ask it to hand a UI task to my frontend agent running Claude Opus. Both can start from the same Checksub context. Here鈥檚 how I set that up.
1
4
I keep my working context and memory in a Git repo so I can switch models without starting over. Tomorrow I鈥檒l share how I set it up and how I decide which tools each AI can access. When you switch AI tools, how do you bring your context with you?
1
7
I鈥檓 back. A few months ago, we changed how we build Checksub with AI. We ship much faster now. I use ChatGPT, Codex, Claude Code and Grok. Talking with other founders reminds me of a workshop. Give two people the same power tool and watch how differently they work.
1
46
Replying to @Bigoudi65Lolo
Il y a 2000 scientifiques qui ont sign茅s une p茅tition d茅non莽ant l鈥檃bsurdit茅 de ce projet. Dont le directeur de recherche du CNRS et des membres du GIEC. Mais cela ne serait pas la premi猫re fois qu鈥檕n 茅coute pas la communaut茅 scientifique.
1
19
Can Elon Musk change the color of the Sun? I'd like to offer you a bit of philosophy to tackle the subject of Elon Musk's considerable influence in the world. Do you think this is normal? Should this influence be limited? By what means? Is it too late?
79
Replying to @icreatelife
@Kris you should try Checksub if you want to translate video content. #aidubbing
7
Replying to @bdemontille
Les gens qui commentent sans lire l鈥檃rticle. Cette d茅cision avait 茅t茅 prise lors du confinement COVID. Le menu v茅g茅tarien est toujours dispo. Et heureusement car c鈥檈st meilleur pour la sant茅 et pour l鈥檈nvironnement. Je vous invite 脿 creuser le sujet.
25
Can I hire you? 馃ぃ
15