How to Build Your First Agentic Workflow with dGENIX
Building an agentic workflow is simpler than it sounds. Define the goal, stack your skills, write the steps, and let GENI run. Here is a step-by-step walkthrough.
Start with a problem, not with technology
Before opening a tool, answer one question: which recurring multi-step task costs your team the most time each week? Good first candidates:
- Writing and sending weekly status reports
- Researching and qualifying new leads
- Handling one category of incoming email
- Turning existing content into social posts
Pick one. The best first workflow solves a real problem, not the most impressive demo.
A good test: can you explain the task in one sentence to a new colleague, including how you know it went well? If you cannot, the task is not sharp enough to automate yet.
Step 1: define the goal and the steps
Write the workflow out as a human would do it. For lead research:
- Take a list of company names
- Research each company: industry, size, recent news
- Find the decision maker
- Write a personal email based on the research
- Put the contact and notes in the CRM
- Schedule a follow-up in 5 days
That is your blueprint: clear input, clear output.
Mark in that list which step is irreversible. In this example that is sending the email; everything else you can throw away and redo. That marking decides where you put approval later.
Step 2: choose your stackable skills
Map each step to a skill:
- Research, Deep Research skill
- Finding contacts, Lead Research skill
- Drafting email, Gmail skill
- Updating CRM, HubSpot skill
- Follow-up, Calendar skill
Activate each skill with one click on the skills page.
Check your connections here too. A workflow with a Gmail step gets no further than that step if your Google account is not connected, and you only find out halfway through the first run.
Step 3: write the instructions
In dGENIX every project folder has its own instructions, and that is where you define the workflow. Tell GENI the goal, the steps and their order, which tool per step, the output format, and decision rules such as skip a company with fewer than 10 employees. The more precise the steps, the more reliably GENI runs them.
Two things that noticeably improve instructions. Describe what the result looks like, not only what has to happen: one example of a good email works better than three lines of style advice. And say what should happen when something fails, for example that a company with no findable decision maker is skipped and listed rather than guessed at.
Step 4: run and review
Run a small batch first, 5 to 10 items. Check that GENI completed every step, that the format is right and that the quality meets your standard. Expect to refine the instructions two or three times. That is normal.
On those first runs, focus on the cases that deviated, not the ones that went well. The five that were fine teach you nothing; the one that got stuck shows exactly which instruction is missing.
Step 5: schedule it
Once it runs reliably, schedule it: daily at 7, weekly on Monday, or triggered by an event. Scheduled workflows run without a human trigger. The output lands in your dashboard, ready to act on.
Watch two practical limits. How many recurring tasks you can have at once depends on your plan. And if your credits run out, a scheduled task pauses rather than half-executing, which you see as a notification.
Where to put approval
Back to that marking from step 1. The rule of thumb: one checkpoint, just before the first irreversible step.
In the lead example that means researching, finding decision makers and writing emails happens unattended, and you see ten drafts staged and ready to send. That is one two-minute action instead of three hours of work.
What you must avoid is approval per step. Then you have built something that demands as much attention as the manual work, and the gain is gone. And what you must also avoid is no control at all on a workflow acting outwardly on your behalf, because then you discover an error when a client replies to an email that made no sense.
After a few weeks you can move that moment or drop it, but do so based on what you have seen, not on trust up front.
Common mistakes
- Goal too broad, manage my pipeline is not a workflow, research 10 leads every morning and put them in HubSpot with drafts is
- Skipping review, run it manually a few times first, edge cases surface early
- Automating everything at once, start with one, run it two weeks, then add the next
A fourth that gets expensive: setting the workflow up and never looking at it again. Systems change, contacts leave, and a workflow that ran well three months ago may quietly be doing something wrong. Book a moment each month to read the latest run.
What to expect
A well-defined workflow saves the average team 3 to 8 hours a week on that process. Three workflows running automatically add up to a full working day less manual work per week.
Do expect the first workflow to cost you time rather than save it in week one. Setting up, testing and refining is half a day to a day. From week two it starts paying back, and on the second workflow you are twice as fast because you know how to write instructions.
Learn the concept behind it in What are agentic workflows, or browse all skills to see which steps you can chain.
