Well I'm always experimenting with new tools. Currently using a custom made orchestrator that also taps into ITSM + Spatie Flare errors and does triaging. Also has custom routines for deployment, etc.
Jérémy Marodon
@Th3Mouk
Principal Product Engineer | Agentic & Harness Engineering | Cloud, Web & DevOps | Building scalable products and empowering engineering teams
- Clermont-Ferrand
- marodon.com
- github.com/Th3Mouk
- Joined October 2026
Commit activity
33 commits in the last year
longest streak 4 days
731 / 1,000 XP269 XP to wheel
Sun, Nov 1
Mon, Oct 19
replying to @nunomaduro
Ahah ! Made the same remark to the team last week. I begun the day with 80PRs opened, and after merging like 25, 100 were to merge 😅 I think it's when you face the hard reality, you are forced to build harness, workflows and skills
Sun, Oct 18
replying to @RaphaelFakhri
💯
At this early point tooling because my harness has been tailored locally for 18 months and now is kinda frozen.
I think the game changer is when you understand that a feedback loop is the key. Each sessions make you more efficient
yeah the feedback loop is the whole game. once every session leaves something behind for the next one, it compounds fast
replying to @RaphaelFakhri
It was too hard to pick a service to me so I picked a VPS and an old computer at home, put a Debian without GUI and here I am.
I finally stopped to think about micro VMs and stuff and scalability. I have automated everything I can and now working on daily to improve rate of…
same setup here honestly, one always-on headless linux box and agents running on it. measuring success rate without steering is the right metric, what's moving it most for you, better prompts or better tooling?
replying to @gnapse
Coding will become a hobby for sure, but we still need people to build products and services. Just with new tools and different responsibilities
replying to @ramzyfayek
PHP coders yes, engineers not at all. The abstraction is moving one level upper. I doubt anyone getting paid to code for a long time, but to build a product, a service, helped by tools like LLMs
hey 👋 harness engineering is my favorite rabbit hole right now. what are you running your agents on?
Sat, Oct 17
replying to @Th3Mouk
And my son burst my Claude quota to design and animate a cat in SVG + Javascript 🤣
replying to @cannonsh
Same. Now that my local harness is totally autonomous with the quality I require, It's time to experiment the final touch of automation
replying to @simonhampel
Congratz 👏 So nice to see theses possibilities. I need to push further my integrations, most of them have crashed today after Tailscale configuration changes. Monitoring/Alerting + SRE agent is the way to go. I had to take a OpenAI account to reduce cost, does it cost much to you…
No API - I'm on subscription, had to upgrade to 20x recently because I'm doing so much coding at the moment, but given how productive it's made me, it's worth every cent.
replying to @MrPunyapal

/dev/nullJérémy (@Th3Mouk) replying to @slovenianGooner on /dev/null: “In fact I hesitated a lot between Debian and Omarchy, but final…”In fact I hesitated a lot between Debian and Omarchy, but final decision was I want a fully autonomous system, so I choose to try going without GUI. Hermes have a dashboard, and Claude and Codex are…
replying to @cannonsh
In fact I hesitated a lot between Debian and Omarchy, but final decision was I want a fully autonomous system, so I choose to try going without GUI. Hermes have a dashboard, and Claude and Codex are able to act throught Tailscale by SSH. It's super easy to maintain and secure
Yeah I am looking to this to have a triage for tickets - putting it into coding tasks.
replying to @IsaacHatilima
The most incredible is the way now session can communicate between themselves, join and quit a task. We no longer need to have a custom CrewAI or LangChain
replying to @IsaacHatilima
The fact being able to switch working dir is madness. No need to put hooks anymore. I've dozens of parallels worktrees, now I can see the code properly and comment it directly into it
replying to @GromNaN
My frequency of usage : Opus 5.5 (brainstorming/exploration/security) > Gpt 6 Luna (all my Hermes agents for daily life, professional work delegation) > Sonnet 5.5 (execution) > Sol 6 (adversial critic/second point of view) > Haiku 5.5 (code exploration, summary, dumb execution…
replying to @IsaacHatilima
Claude Desktop is just what I ever wanted and tweak around for months. Last version is perfect I no longer need to setup anything or open an IDE