support should install itself
4 min read
we are in 2026, and customer support still feels strangely manual.
DunSocial is growing, so recently i started looking for a support tool. my goal was very simple: i don’t want to manage support tickets.
most questions should be handled by agents. not drafted by an agent and sent to me for approval. not categorized by an agent so i can open the inbox later. actually handled by the agent from beginning to end.
but setting up these tools still means doing a lot of the work yourself.
create the inbox. install the widget. write the help articles. upload the knowledge base. organize it. keep it updated. tweak the bot. connect every action manually. then keep checking the tickets the ai could not resolve.
that is the frustrating part.
if i have to explain my entire product to the support tool, maintain another version of my documentation, and still manage the queue, what exactly is the agent doing for me?
the here.now idea
recently i looked at here.now, and i loved how simple the idea was.
there is an instruction on the landing page. you copy it, give it to your agent, and the agent does the rest. it can deploy the html page, upload the files, create the storage, and give you a live url. you don’t need to create an account before doing the thing.
if you want to own it later, you claim it with a token. if nobody claims it, it eventually expires.
that flow is so straightforward.
copy the instruction. give it to the agent. get the result.
no long onboarding. no dashboard tour. no twenty-step setup before you understand whether the product is useful.
it made me wonder why customer support cannot work the same way.
give the agent one instruction
imagine copying one system instruction into your coding agent.
it reads the codebase. it understands the product, the routes, the plans, the features, the billing rules, the docs, and the recent changes. it creates the knowledge base by itself. it deploys the support page. it adds the widget. it connects the right tools. it keeps everything in sync as the product changes.
you don’t upload anything.
you don’t rewrite your product into help-center language.
you don’t maintain the same knowledge in five different places.
it should already be there in the codebase. the agent should do the work of understanding it.
then i should be able to give it a few custom actions:
- refund an eligible payment
- upgrade or downgrade a plan
- check billing and usage
- restore access to an account
- cancel a subscription
- escalate the few cases that actually need me
this is what i want from an agent-first support tool. not another chatbot sitting on top of a ticketing system. i want an agent that can understand the problem, take the action, verify that it worked, and reply to the customer.
the default should be resolution, not escalation.
ai support is still built around tickets
most tools added ai to the existing support workflow.
that makes the replies faster, but the basic idea is still the same: a customer creates a ticket, the ticket enters a queue, and eventually a human owns it.
agent-first support should reverse that.
an agent owns the request by default. a human only enters when there is real ambiguity, risk, or something the agent is not allowed to do.
maybe this is not the ideal solution for hundreds of companies with complicated policies, permissions, and support teams. i understand why the existing tools are built the way they are.
but i am not trying to solve that first.
right now, i want to solve it for myself.
DunSocial needs support. i don’t want to build a support operation around it. i want to give an agent access to the product, define the boundaries, and let it take care of the rest.
we are in 2026. agents can read an entire codebase, build a product, deploy it, and fix it when something breaks.
but to set up customer support, i still have to manually write the knowledge base.
that feels backwards.
support should install itself. learn by itself. resolve things by itself.
and only call me when it truly needs me.