Everyone is building towards each other: how AI agency workflow automation works in the gaps

The conversation that made this obvious

Stefan Maritz, founder of Contengi, put it plainly in a recent field note: "Everyone is all of a sudden trying to figure out how to build towards each other, how to plug in one MCP or API. Can I do this? How can our systems speak to each other to automate a flow or build something between the two companies that has never existed before?" That observation landed during a live conversation with a PR agency, and it lands the same way every time.

The most interesting AI agency workflow automation happens when an agency's data layer connects to a client's knowledge base and a third-party signal feed. That junction - where systems connect - is where the real infrastructure is being built right now.

What this actually looks like: the PR agency example

The PR agency conversation had two concrete workflow ideas on the table. Both show the principle in action.

The first idea: pull the PR agency's data directly into the client's AI setup - what Stefan calls the GTMO, a generalised term for the content operations environment - so that instead of logging into separate dashboards and manually pulling reports, you can query your AI assistant directly. As Stefan described it, "I could just ask my assistant, my AI assistant, 'Tell me what impact the last six months of media had - the PR - and if it's being cited in LLMs, and is it giving us any traffic or brand uplift or backlinks or whatever the case may be?'" That single query crosses PR data, LLM citation signals, and Google Analytics in one ask.

The second idea is more interesting. The PR agency's AI system scans the market for opportunities - topics, angles, journalist questions gaining traction - and sends a weekly report to the client. The client then runs those opportunities against their own knowledge base: transcripts, podcast conversations, internal discussions. "If there's a conversation happening around whether banks are ready for methos, do we have a transcript that already addresses this, either internally or from the podcast or with a partner?" If yes, automation can start drafting from that transcript immediately. If there is no transcript covering the topic, the system flags it and sends the podcast programme manager an email to prompt a new conversation.

What changes when automation crosses organisational lines

Standard AI workflow automation content focuses on what happens inside a single tool: how an agent routes a support ticket, how a platform auto-generates a report, how a chatbot handles a query. That is fine, but it is a narrow frame. It treats the system as a closed loop.

What the PR agency example describes: value lives at the handoffs. The PR agency has market signals the client does not have. The client has proprietary knowledge - conversations and transcripts, subject matter expertise - that the agency cannot access. When those two data sets talk to each other through a shared workflow, you get something that neither party could produce alone. As Stefan put it: "I've got X, you've got Y, how can we put the two together and make something incredible together in workflow? Because not everyone is gonna build everything, but together we can build anything."

IBM describes agentic workflows as AI-driven processes where autonomous agents coordinate tasks with minimal human intervention. The PR agency model described here is an applied version of that - two organisations' agents coordinating across an API boundary to produce outputs neither could generate solo. The Content Marketing Institute's work on agentic content workflows points in the same direction: the power is in connecting narrowly focused agents with clear roles and shared context.

The knowledge base as the connective tissue

The knowledge base on the client side is what makes this whole architecture work. If the PR agency sends over a market opportunity and the client's system has no structured knowledge to run it against, the workflow stalls. The opportunity sits in an inbox and someone has to manually figure out whether the brand has anything useful to say.

When the knowledge base is properly built - with transcripts, podcast episodes, partner conversations, internal expertise all indexed and queryable - the match-making becomes automatic. The system can surface a relevant transcript and start drafting from it, producing content that is grounded in original thought and proprietary insight drawn from real conversations. That produces non-commodity content that reflects the brand's expertise.

Setting up that knowledge base properly is not a small task, but it is the foundational investment that makes every downstream workflow more powerful. With a properly built knowledge base, every downstream workflow automates real substance. There are ways to structure this so agents can use it - the setup decisions matter more than the tooling. Contengi's AI knowledge base setup guide is a good place to start.

The newsjacking loop that runs itself

What Stefan described in the PR agency conversation is, at its best, a fully automated newsjacking loop. Journalists ask questions. The PR agency's AI captures those signals and passes them on. The client's system checks whether it has relevant expert knowledge. If it does, a draft begins. If it does not, the programme team gets a prompt to go create it.

That loop - signal in, knowledge check, draft out or gap flagged - is a media system. One that does not depend on a human sitting in a monitoring dashboard at the right moment, or a content manager manually reading PR reports and connecting them to editorial calendars. The system does the connecting. Humans step in for judgement, approval, and quality.

This is where agentic content workflows get genuinely interesting for agencies and their clients. The value lies in the chain of decisions the system makes across organisational boundaries, decisions that used to require multiple people in multiple meetings to coordinate.

What it takes to build towards each other

Two things make this kind of cross-system workflow possible. First, both parties need structured, queryable data - the PR agency needs its market intelligence in a form an API or MCP can serve, and the client needs a knowledge base that a query can search. Second, there needs to be a shared protocol for how the data moves: a webhook, an API call, an MCP server, something that lets the two systems exchange signals without a human acting as the intermediary.

This is what Stefan means when he talks about the average agency starting to think in systems: "It's not enterprise solutions that are trying to tailor make to you. It's the average agency with the platform at the moment that they build - everyone is just starting to think, what can AI do?" The questions being asked are practical ones. What data do you have, and what happens if we connect them? Those questions are more valuable than most of the strategy decks being written about AI right now.

For anyone running agentic content workflows for a small business, the principle is the same even if the scale is different. Your podcast guests, your client conversations, your internal expertise - that is your X. Your agency partner's market signals or audience intelligence is their Y. The question is whether you have built a system that can receive the handoff.

Start with the question, not the tool

The instinct when reading about this kind of infrastructure is to go looking for the tool that makes it happen. The right starting point is the question Stefan asked at the top of that PR agency conversation: what data do you have, what can my system do with it, and what can we build together that neither of us can build alone?

That question, asked between an agency and its clients, is producing workflows that did not exist eighteen months ago. It is producing media systems that self-populate and editorial calendars that respond to live market signals, all grounded in genuine subject matter expertise rather than whatever the AI can hallucinate on demand. The collaboration question precedes tool selection, always.

Start by looking at what data you already sit on - your transcripts and conversations, your proprietary knowledge - and ask which of your partners has the signals you are missing. That is where the workflow begins.

Frequently asked questions

What is AI agency workflow automation and how is it different from standard automation?

AI agency workflow automation uses AI agents to move data and decisions across systems - including across organisational boundaries between an agency and its clients. Standard automation follows fixed rules inside a single platform, while agentic workflows can interpret context, make decisions, and coordinate across multiple tools and data sources.

How do MCP and APIs enable cross-agency AI workflows?

MCP servers and APIs give two separate systems a protocol for exchanging data without a human acting as the go-between. An agency's AI can push market signals to a client's knowledge base, trigger a query, and receive a response - all without anyone opening a dashboard or writing a report manually.

Why does a knowledge base matter so much in agency AI workflows?

The knowledge base is what gives the client's system something to match against when an agency sends over a market signal or opportunity. Without structured, queryable knowledge - transcripts, conversations, subject matter expertise - the workflow has nowhere to go and a human has to step in to do the connecting manually.

Can small agencies and their clients build this kind of cross-system workflow?

Yes, and it is increasingly common. The PR agency example in this piece is a small-to-mid-sized operation, not an enterprise with a dedicated AI team. The foundational requirement is structured data on both sides and a shared integration point - an API, a webhook, or an MCP server - that lets the two systems talk.

What is the first step in building an AI workflow that connects agency data to client systems?

Start by mapping what data each party already holds in a structured format. Then identify one high-value signal - market opportunities, PR links, citation data - that the agency can push to the client system on a regular cadence. Build the integration for that one signal first, validate it produces useful output, and expand from there.