I'm still struggling with setting up long running agents. So far, it's just summoning an agent with a skillset for something specific and then it leaves again. Any tips?
Munder Difflin – Agent harness to run an office of your clones
81–90 of 162 posts
Re: Munder Difflin – Agent harness to run an office of your clones
#82[flagged]
Re: Munder Difflin – Agent harness to run an office of your clones
#83Earlier quoted context omitted.
It is nice to see folks experimenting widely with unique visual representations of agent orchestration, as a consequence of the bottom-up / individual-led development that's driving the industry (inasmuch as it exists yet). There are folks young enough now that they've never known anything other than "the desktop" (or CLI) as a computing metaphor. But there's no fundamental reason a metaphor has to be anything specif…
LLMs remade Clippy. This sounds like using Microsoft Bob to manage them.
Imagine how much more infuriated people would have been if Clippy wasn't a cartoonish-looking paperclip, but a human facsimile.
Sometimes absurdity or comedy is the point when you're developing a known-partial solution.
Re: Munder Difflin – Agent harness to run an office of your clones
#84I do like the embrace of The Office as a theme because it so accurately represents the dysfunction of all of the agent swarms I’ve seen: Different personalities pursuing their own little goals that are all competing with each other in subtle ways that eventually leads a funny collapse of the outcome you wanted.
There’s no denying that LLMs are getting better by the month, but the current wave of LLM office and personal assistants reminds me of the old trend where people were hiring personal virtual assistants from foreign countries to manage their work and interface with people. It seemed like an obvious time saver but every time I interacted with someone’s virtual assistant it felt like I was playing a little game of navigating hidden structures and communication barriers to get the message I needed to the person hiding behind it all. Then there were the inevitable scheduling failings, missed meetings, dropped emails, and other things that get blamed on the assistant. I’m getting deja vu watching it all happen again with LLMs replacing virtual assistants and outsourced teams.
Maybe these concepts work for people who are trying to solo dev and who find it interesting to set up and debug little systems for everything they do, but I really don’t like working with anyone who surrounds themselves with one of these multi-agent coordination systems as an external shell. Keep it to the internal work and maybe it’s fun for some people.
Re: Munder Difflin – Agent harness to run an office of your clones
#85Ok, I've been running it for a couple hours and below are my thoughts. Please note that I do find it fascinating even if most of what I'm about to say is complaining about the parts I like less. - Pipelines, not agents. Roles, not agents. I really don't like the idea of defined agents with their own prompt. I want to define roles and spin up N agents with that role. Furthermore I want pipelines "Plan -> Review Plan -…
Pipelines and roles are a good point.
Re: Munder Difflin – Agent harness to run an office of your clones
#86Hey guys, thanks for putting it here, I am Chaitanya I built Munder Difflin, I am here to answer all your questions(except nylonstrung). For people who haven't tried it: Munder Difflin is a local multi-agent harness that wraps around your existing claude code and codex subscriptions(we literally support almost all harnesses/coding agents). Simulations are deterministic, they do not consume tokens, infact most of the…
when did you launch, how many users you have, and what's the business metrics if you don't mind sharing.
Re: Munder Difflin – Agent harness to run an office of your clones
#87I've thought about something like this. When there is a lot of stuff happening at once, it's a mistake to try and communicate it all with text. Agents use tools, reference databases, reference the web, interact with other agents, and spend time processing the information. When you have several agents operating at the same time, communicating what they are doing using some kind of spatial map is a really smart idea. T…
You are correct this is interesting but it is still quite obviously the wrong UI. For the same reason https://www.gather.town/ ends up being a waste of time.
It's been done before with LLMs. In both cases - human and LLM - trying to "visualize" what people are working on by having a topdown office view is a complete waste of time. It would be like thinking by strolling into a random office IRL and surveying people with their heads down at a random moment you will unlock some great insight.
It's cute. But it's just that. If you want to actually understand some of the things you highlight here there are far more accessible and efficient ways. Like take any data plot or visual graph.
An office is simply the wrong map. It also misses the nature of the joke in The Office. No one is really getting work done. Their work is boring and sucks, no one cares. The whole show is essentially the domino effect of Michael as a rouge agent. They have many scenes where a new character comes in and sees the whole office. It appears normal. Yet after this moment some chaos quickly begins to brew. My point is unless you are planning for a Michael to go haywire, the entire concept of an office is completely uninteresting and tedious. You may as well just use a spreadsheet.
Re: Munder Difflin – Agent harness to run an office of your clones
#88Hey guys, thanks for putting it here, I am Chaitanya I built Munder Difflin, I am here to answer all your questions(except nylonstrung). For people who haven't tried it: Munder Difflin is a local multi-agent harness that wraps around your existing claude code and codex subscriptions(we literally support almost all harnesses/coding agents). Simulations are deterministic, they do not consume tokens, infact most of the…
“send cold emails with enriched context” What’s the difference between cold and spam?
I often get cold emails from people interested in my very specific hobby (see website in profile). These are desired and fun.
Mind, they are not written by LLMs. I would consider that spam, outside of very limited circumstances (maybe a non-english speaker using an LLM for translation).
Re: Munder Difflin – Agent harness to run an office of your clones
#89Hey guys, thanks for putting it here, I am Chaitanya I built Munder Difflin, I am here to answer all your questions(except nylonstrung). For people who haven't tried it: Munder Difflin is a local multi-agent harness that wraps around your existing claude code and codex subscriptions(we literally support almost all harnesses/coding agents). Simulations are deterministic, they do not consume tokens, infact most of the…
“send cold emails with enriched context” What’s the difference between cold and spam?
Nailed it. I constantly get “cold emails” that were clearly written by an AI that read 2-3 pages on my web site, my LinkedIn bio, and something else on my social media. They all follow the same formula: try to break they ice with something funny about somewhere I took a photo recently, then segue into what I do for work, then offer to help me do it better - usually by offering leads or CRM improvements.
It’s such an obviously repetitive recipe that someone must have published as an example somewhere, and now lots of people (or bots) are trying to follow it, all mining for gold that doesn’t exist.
Re: Munder Difflin – Agent harness to run an office of your clones
#90hey neat stuff. I'd actually like the option for the simulation NOT to be deterministic. I don't mind sacrificing 100% of any productivity for this.
1. Pranking: Rejected my PR? I'll put your keyboard in jello. Then the jello'd agent's prompts all get your keyboard is currently in jello attached.
2. Office Romance: Certain agents prefer to work with each other, but can randomly experience entertaining breakups.
3. Dundies style award ceremonies. The titles can show up under their names until the next ceremony i.e. "Hottest Agent in Office"