Agentic Workflows
Agentic workflows that do a job, not a demo
I build one production workflow at a time, with monitoring, guardrails and a proper handover, on the same pattern I use to run my own businesses.
What an agentic workflow is
Not a chatbot. A piece of work that gets done.
An agentic workflow is a process where an AI agent works through a series of steps: it reads the input, decides what to do, uses tools, writes the output and checks it against rules you set. People sign off at fixed points. Nobody has to prompt each step.
Compare that with a chat tool. A chat tool helps one person do a task. A workflow takes the task off the list.
Where they pay off
Four places I see the fastest payback
Content.
Search data picks the topics, an agent drafts and structures each post, a checker enforces your house rules, and a person approves before it publishes. I run this on my own sites every week.
Operations.
Data syncs between systems, exception reports and routine updates. The agent does the re-keying and flags the odd ones out.
Sales.
Research on inbound leads, first-draft replies, follow-up sequences and CRM hygiene, with a person owning every send that matters.
Data and reporting.
Weekly reviews that pull the numbers, spot what changed and write it up in plain English, so you read the conclusion rather than the spreadsheet.
Want the content workflow run for you every week? See GTM engineering. Want AI assistants to find and quote your business? See AAIO.
What you get
One workflow, in production, that your team can run
- A value-stream map of the process, with the waste removed before anything is automated
- The workflow itself, built into the tools you already use where that's possible
- Monitoring, so you know when it runs, what it did and when it fails
- Guardrails: written rules, checks on the output and a human review queue for anything risky
- A handover: runbook and a walkthrough with the people who'll own it
4–8 weeks. Fixed scope, typically £8,000–£25,000.
How a build runs
Five steps, in this order
- 01
Map.
Walk the process as it runs today and time it.
- 02
Cut.
Remove the waste. Some of it never needed automating.
- 03
Split.
Run the Knowledge · Skills · Wisdom test on every remaining step: AI, agent skill, or person.
- 04
Build and guard.
Build the workflow, the checks and the monitoring together, not the checks as an afterthought.
- 05
Hand over.
Your team runs it for two weeks while I watch, then it's theirs.
What I won't do
A few things I'll say no to
- Build an agent on top of a process we haven't fixed
- Hand over something only I can maintain
- Let an agent publish, send or pay anything with real risk attached without a person signing off
- Quote you a number I can't back up
Proof from my own production lines
Workflows I run every week
Campercation
A campervan travel platform I own and run
The blog runs on a search-seeded pipeline: topics come from search data, drafts are generated and GEO-structured, images go to a CDN and posts publish through a .NET API. A weekly SEO loop decides what to write or update next.
25,000+ stopovers on the live map
RunningTrustedVillas
A villa rental platform I own and run
A property sync pipeline with monitoring, so listings stay current without anyone re-keying them. Hub content sits under change control, so an agent can’t quietly rewrite a live page.
63,000+ properties synced
RunningVerqos
The content engine behind both
The same pipeline, productised. It plans, drafts, checks and publishes, with a human review queue for anything legal, pricing or competitive.
Hundreds of hours saved
RunningHow I run the business
Scheduled agent tasks
My weekly LinkedIn and SEO reviews run as scheduled agent tasks with written quality guardrails. I read the output, not the raw data.
24/7 automated agents running
Running
FAQ
What is an agentic workflow?
It's a process where an AI agent carries out several steps on its own, such as reading, deciding, using tools and writing, with checks built in and people signing off at set points. The point is that the work gets done without someone driving each step.
What's the difference between an agentic workflow and automation?
Traditional automation follows fixed rules: if this, then that. An agentic workflow can handle inputs that vary, like an email, a document or a messy spreadsheet, and decide the next step. The best builds use both: fixed rules where the path is known, and an agent where judgement within limits is needed.
What are some examples of agentic workflows?
A content pipeline that picks topics from search data, drafts, checks and queues posts for approval. A property sync that keeps listings current across systems and flags errors. A weekly review that pulls your numbers and writes up what changed. I run all three on my own businesses.
How long does it take to build an agentic workflow?
A single production workflow typically takes 4–8 weeks, including mapping the process, building, testing and handover. Simple ones are faster. Most of the time goes on the process and the guardrails, not the AI.
What happens when the agent gets it wrong?
It will, sometimes, so the build assumes it. Outputs are checked against rules, anything risky goes to a person, and monitoring tells you when a run fails or drifts. You see the error before your customer does.
Who owns the workflow after handover?
You do. It runs in your accounts, your team has the runbook and the walkthrough, and you're not tied to me to keep it running.
Got a process that eats someone's week?
Tell me what it is. I'll tell you straight whether an agent should do it.