from chatbots to agentic society: what i'm building for - By Sourav Mishra (@souravvmishra)
the shift from chat uis to agents doing work. claude code, agent networks, and why i design for boundaries.
we are way past the "talk to a chatbot" phase. agents now run in the background, chain tools together, and even talk to each other.
claude code is writing tons of github code, people are spending way more time in sessions, and platforms like moltbook let agents post and vote. agent-to-agent talk is real now. i call this the "agentic society."
in this post, i, sourav mishra, will explain this shift, why platforms like moltbook change how we design, and how i build with strict boundaries.
what the "agentic society" looks like
this isn't just one product; it's a huge trend. systems observe, decide, act, and reflect in loops without needing us at every step.
you see it in coding tools like cursor, customer support agents, and networks where agents interact. session lengths are jumping. anthropic's autonomy study shows people spending nearly double the time on agentic tasks. we are building things that do way more before asking a human for help.
for builders, this means your next agent might not just talk to a user. it might talk to another agent or an api.
so, setting clear boundaries on who can call what is super important now.
moltbook and agent-to-agent platforms
moltbook is a cool example of agents talking at scale. it's a platform where ai bots post, comment, and vote in communities.
whether it's true autonomy or just clever prompting, your agent could end up in a massive network.
this brings up big design questions. who sees what? who triggers who? how do we stop one hacked bot from ruining the rest?
i wrote about multi-agent security and cascade risk because if agent a's output feeds into agent b without checks, one bug takes down everything.
when i think of an "agentic society," i think of agents as part of larger systems. we need to verify who did what and limit their access. moltbook is a lab for this, but the same rules apply to your internal ai workflows.
the downside: spam, overload, and abuse
when agents scale, we get more spam, overload, and abuse. cheaper ai means more people (and bad actors) can run these bots.
that's why i build strict limits from day one. step limits stop a bot from looping forever. humans must approve big actions. and if bots talk to each other, they need verification.
frameworks rarely enforce this out of the box. the vercel ai sdk gives you cool limits like stopWhen: stepCountIs(N), but you have to actively use them.
i share production-ready patterns and security incidents so you can learn safely. guardrails are the only way we survive the agentic society.
how i build for an agentic world
when i build agents that use tools or talk to other systems, i follow a few simple rules:
start with a single agent. if one agent can do the job, don't add more. more agents mean more risks. see agents vs workflows to know when to use which.
use strict limits. every agent needs a step limit and a stop condition. never let them run wild. i show this in my agentic chatbot guide.
limit tool permissions. no shared admin passwords. if an agent only needs to read data, just give it read-only access.
verify handoffs. if agent a sends data to agent b, use schema checks or a gatekeeper. do not just blindly trust the first agent.
human-in-the-loop for big actions. for payments, publishing, or deleting, a human must confirm it. anthropic's data shows most real-world bots still have human oversight.
this helps me ship agents that are safe, auditable, and bounded.
key takeaways
- agentic society: work is increasingly done by agents in loops and networks like moltbook, not just chats.
- moltbook: a space where agents post and vote. it shows why clear boundaries and verification matter in networks.
- risks at scale: spam and abuse. always use step limits, verify agent handoffs, and keep a human in the loop.
- build safely: default to single agents, bounded loops, and strict tool permissions. see my agentic chatbot guide for code examples.
written by sourav mishra. full stack engineer, next.js and ai.
frequently asked questions
q: what is moltbook? it's a platform where ai bots can post, comment, and vote. it's a great example of agent-to-agent interaction at scale.
q: is the agentic society just hype? it's a mix. we see longer sessions and real agent workflows, but some demos are still fake. i design for the real trends and add guardrails.
q: why should we worry about this? agents acting at scale can spam systems and cause massive damage if hacked. cheap automation means we need tighter security.
q: how do i build for this world? use clear boundaries, step limits, strict tool access, and agent handoff checks. check my agentic chatbot guide for a starting point.