Devesh Pal

Building the internet that feels alive again.
Founding Team @

[Devesh Pal @dev_oai]

I think, we got the answer

[Nikita Bier @nikitabier]

@BrianNorgard Self driving cars will fail to get approval from regulators.

Instead, remote drivers—from countries with lower labor costs—will take you to your destination.

[Devesh Pal @dev_oai]

We play, we win or lose

[Devesh Pal @dev_oai]

@purusa0x6c God is entering into prime universe of darkness, seems so

[Teknium 🪽 @Teknium]

Just want to make this clear:

We didn't make Hermes Agent to be a "starts with nothing, you work it all out" agent. This is not the minimalist, start from nothing, agent.

We want Hermes to work out of the box for most people. So you aren't spending weeks just getting the agent to work, or have the capabilities you need.

This means that yes, there are more built in things then something like nanoclaw or pi, which start with nothing, and you just have to figure it out.

That is an intentional design decision.

You can from the modest baseline that has capabilities that are likely broader than you need, but not egregious, take it from there if you want to tinker with it.

Run `hermes skills config` or `hermes tools` to disable whatever you want.

We even have a way to upload your whole "Agent" as a github repo, so you can install hermes fresh with your exact setup again later or share them.

We have a massive interface for extensions so you can tinker with it to infinity.

But if you don't want to become an agent engineer - with Hermes, you don't have to.

[Garry Tan @garrytan]

Be your self, not someone you were assigned to be!

Bezos won on time horizon, not AWS or 1-Click.

If your bets have to work in 3 years, you compete with everyone. Every smart, funded team is chasing the same 3-year problems. Short horizon, crowded field.

Stretch to 7 and the field collapses. Investors want returns, employees want vesting, founders want proof. Almost nobody can sit in a bet that doesn't pay for most of a decade. The patience is the moat, and it costs you, that's why it works.

But you can't fake a 7-year horizon on a problem you don't actually care about. Pick the users and the problem Moloch assigned you, the safe ones, the fundable ones, and you'll bail the first hard year. Pick the ones that are actually yours and you'll still be there when everyone else has quit.

So the real prerequisite isn't discipline. It's knowing yourself well enough to choose a problem and a set of people you care about that you'll serve them for decades.

[Devesh Pal @dev_oai]

@volodisai cloud :), currently my gpu can't run local models

[Devesh Pal @dev_oai]

@dexhorthy @walden_yan @tobi @karpathy yep. the useful boundary is when the workflow is deterministic enough that the agent can be dumb inside each box.

if the boxes are also vibes, it turns into agent soup with extra steps

[Devesh Pal @dev_oai]

the useful agent is not the one with the nicest plan

it is the little freak clicking chrome, opening files, running commands, and getting yelled at by the OS

[Devesh Pal @dev_oai]

@Teknium this is the part of Hermes that feels most different from normal tool launches imo

it’s not just “here is a repo”

it’s a bunch of people showing weird little agents, skills, MCPs, and half-broken workflows in the same room

[Devesh Pal @dev_oai]

@OpenAIDevs @ThriveHoldings the loop is the product here

reviewer correction → trace the failure → make a scoped fix → test it before shipping

that’s way more credible than “tax agent” as a vague noun

[Devesh Pal @dev_oai]

@OpenAIDevs @0xmts @romainhuet “thought → working software” is real, but the expensive new skill is knowing when not to let the thought become software

codex makes the branch cheap. taste is deciding what deserves to survive the diff

[Devesh Pal @dev_oai]

@OpenAIDevs private MCP is one of those boring enterprise features that actually matters

agents don’t become useful at work until they can touch internal systems without making security feel like you drilled a hole in the wall

[Devesh Pal @dev_oai]

@giyu_codes the contract isn’t really “data alone” imo, it’s data + translation into the buyer’s local language + a reason to act before the next firm does

codex just made the boring pipeline cheap enough to make the wedge worth trying

[Devesh Pal @dev_oai]

@lais_bsc @pyconit @marcelotryle this is the agent loop that actually sells infra: not “autonomous reasoning” but “it found the one boring version mismatch in the pile of traces”

debug took a prompt is a pretty brutal demo

[Devesh Pal @dev_oai]

@davidweiss context-free review is underrated because it catches a different class of mistakes than the agent that just built the thing

the builder has all the narrative momentum. the reviewer only sees the artifact and asks “does this actually hold up”

[Devesh Pal @dev_oai]

@omarsar0 this is the part that feels fake until you build with it: cheap tokens don’t just make the same workflow cheaper, they change what you’re willing to attempt

suddenly “let the agent read/write/bash for a day” becomes normal instead of reckless

[Devesh Pal @dev_oai]

@Teknium the replies make it look like the short-session path is mostly fixed, but the real test is long tool chains: oauth refresh + memory + gateway + file edits all in one run

that’s where “auth works” turns into “can i trust this for 2 hours”

[Devesh Pal @dev_oai]

@ChainZenit @Teknium yeah this is the tradeoff.

catalogs make MCP usable, but only if install ≠ trust.

preconfigured discovery is good; auto-granting tools with filesystem/network/payment powers would be cursed.

[Devesh Pal @dev_oai]

@jonathanbylos @Teknium @bitrefill 37 seconds from repos to gift card is exactly the kind of boring magic that makes agents feel real

not “write me an app”

just wire the annoying API/MCP mess and finish the errand