A practical field guide from Automation Ace.
Short answer: to go from prompts to processes with Viktor, start by asking it to do a job in Slack or Teams, correct it a few times until the output is right, then have it written down as an editable procedure (Viktor calls these skills and can learn them from a screen recording). Turn the procedure into a scheduled task, which stays paused until a person approves it. Set guardrails so sensitive actions need approval or are forbidden, and review results and credit usage regularly. The result is a recurring workflow that runs on its own but stays readable and supervised.
Feature names and behavior reflect Viktor's materials at the time of writing.
Short answer
- Connect the apps the job needs, nothing more.
- Do the job together by prompt; correct the output.
- Save it as a written procedure you can edit.
- Approve a scheduled task to run it on its own.
- Require approval for sensitive steps; forbid risky tools.
- Review outputs and credits; refine the procedure.
From prompt to process: the five stages
| Stage | What it looks like | Your role |
|---|---|---|
| 1. Prompt | You ask once in a channel: “Pull this week's signups and summarize trends” | Check the result |
| 2. Repeat with feedback | You ask again, correcting format and sources | Shape how it should be done |
| 3. Procedure (skill) | The job is written down as an editable procedure | Read and edit the procedure |
| 4. Scheduled task | The procedure runs on a schedule, after you approve the task | Approve the schedule; review outputs |
| 5. Supervised autonomy | It runs on its own; sensitive steps still wait for approval | Approve exceptions; audit regularly |
Move up one stage at a time, per job. For the general autonomy ladder, see what to delegate to an AI agent.
Step 1: Connect only the tools the job needs
- Connect the CRM, project tool, or docs the job reads from.
- Prefer read access first; add write access only when needed.
- Keep sensitive integrations private to the people who should use them.
Viktor's own advice is to start with a few integrations rather than everything at once.
Step 2: Do the job together a few times
- Ask in a shared channel so the team sees the work and the corrections.
- Be specific about sources, format, and where the result goes.
- Correct it in the thread (“use last full week, not the last 7 days”).
- Stop when two or three runs in a row need no corrections.
Step 3: Turn it into a written procedure
Have Viktor write the job down, or show it once with a screen recording and let it draft the procedure. Then edit it like a handoff document for a new hire:
Procedure: Monday pipeline review (example)
Goal Give sales leadership a 5-minute view of pipeline health every Monday.
Sources CRM: open deals, stage, amount, last activity, owner
Calendar: meetings booked last week
Steps 1. List deals with no activity in 14+ days, grouped by owner.
2. List deals that moved stage last week.
3. Compare total pipeline with the previous Monday.
4. Write 3 bullet insights and 3 suggested actions.
Output Post in #sales-leadership as one message with a table.
Rules Read-only in the CRM. Do not email customers.
If CRM data looks incomplete, say so instead of guessing.
Schedule Mondays 8:30 a.m. local time (approved by: Sales Ops)
Because the procedure is plain language, anyone can see what the AI will do and fix it, which is harder with a large visual workflow. Writing tips: how to build your first AI agent.
Step 4: Schedule it, with approval
- Ask Viktor to run the procedure on a schedule; it suggests a scheduled task that stays paused.
- A person reads the task and approves it before anything runs.
- Pick a sensible cadence: each run uses credits, so a daily task costs more than a weekly one.
- Name an owner for each scheduled task.
Step 5: Set guardrails for autonomy
| Control | Use it for |
|---|---|
| Approval cards (Approve / Reject) | Sending, posting externally, updating records, spending money |
| Tool settings: needs approval or forbidden | Locking down risky actions per integration |
| Who can use Viktor and who it can message | Limiting access in large workspaces |
| Private integrations | Keeping a sensitive connection to the people who should use it |
| Paused scheduled tasks until approved | Making sure nothing recurring starts without a person agreeing |
How approvals work day to day: human-in-the-loop automation with Viktor. General framework: permissions, approvals, and audit trails.
Step 6: Review, measure, and improve
- Spot-check outputs weekly at first, then monthly.
- Track time saved, corrections needed, approvals, rejections, and credits per run.
- Edit the procedure whenever you correct the output twice for the same reason.
- Pause tasks that nobody reads.
See testing and monitoring AI agents and what an AI agent costs to run.
Good first autonomous workflows
| Good first autonomous workflows | Why |
|---|---|
| Weekly pipeline, support, or marketing reports | Read-only, recurring, easy to check |
| Daily triage summary of a shared inbox or channel | Saves time; a person still acts |
| Research briefs before meetings | Varied inputs, clear output |
| Data cleanup suggestions in a CRM or sheet | AI suggests; a person or rule applies |
| Drafting recurring updates (investor, client, team) | Drafts go through approval |
When to hand off to Zapier or Make
If a procedure becomes fixed, high-volume, and rule-based (every new form creates a record and sends a templated email), move that part into Zapier or Make, where it is cheaper and fully predictable, and let the AI employee handle only the judgment. See AI employees vs traditional automation and AI agents vs automations.
Frequently asked questions
How do I build autonomous AI workflows with Viktor?
Ask Viktor to do the job in Slack or Teams, correct it until the output is right, save it as a written procedure, approve a scheduled task to run it on its own, require approval for sensitive actions, and review outputs and credit usage regularly.
What is a Viktor skill?
A skill is a procedure Viktor has learned, written down as a document you can read and edit. Viktor can draft one from a screen recording of you doing the job once.
Do Viktor scheduled tasks run automatically?
Once approved, yes. Viktor suggests a scheduled task and leaves it paused until a person on the team reads and approves it. Each run uses credits.
What should I automate first with an AI employee?
Start with recurring, read-heavy jobs that are easy to check, such as weekly reports, triage summaries, meeting research briefs, data cleanup suggestions, and drafts that go through approval.
When should a Viktor workflow move to Zapier or Make?
When a step becomes fixed, high-volume, and rule-based. Move that part into Zapier or Make for predictability and lower cost, and keep the AI employee for the judgment.
Disclaimer: Viktor’s features, integrations, and credit pricing change, and many product details come from Viktor’s own materials; details reflect public documentation and coverage at the time of writing, so confirm current details with each provider before building. Automation Ace is not affiliated with Viktor beyond the referral links marked on this page. This article may include links to apps, products, or services; some links may be affiliate links, which means Automation Ace may earn a commission at no extra cost to you.