The Solo Translator’s AI Workflow Part I
This guide documents a working multi-agent AI system designed for solo translation professionals. It is not a prescription. It is a starting point that you should adapt to your own practice, your own clients, and your own tolerance for automation.
The core question this framework addresses: how does a single professional separate high-value work from low-value operations without hiring staff, without compromising client confidentiality, and without losing control of the decisions that matter?
For the narrative behind this framework — why it was built and what the daily experience looks like — see I Stopped Searching for Translation Jobs. Now They Find Me Every Morning. on our blog.
The agent hierarchy
The system uses three functional roles. You may implement them with different tools; what matters is the separation of responsibilities.
Role 1: Research assistant (isolated)
Mission: 24/7 automated data gathering from the open web.
What it does:
- Scrapes job boards on a regular schedule (e.g., every 6 hours)
- Monitors equipment and software pricing for procurement decisions
- Tracks competitor activity and industry publications
- Feeds all results into dedicated communication channels
Critical security architecture:
This agent runs on a separate, dedicated machine — an older laptop repurposed for this single function. The isolation is deliberate and non-negotiable:
| Has access to | Does NOT have access to |
| The open web | Your working/personal laptop, files, projects or data |
| Its own dedicated email address | Your email address(es) |
| Outbound posting to communication channels (I choose restricted dedicated Slack channels; it does not have access to Slack) | Cloud storage (Google Drive, Dropbox, etc.) |
The data flow is one-directional: the research assistant pushes results outward to communication channels. Nothing flows back. No client information, no project details, no credentials exist on this machine.
For professionals handling legal, humanitarian, or corporate documents under confidentiality obligations, this isolation is not a convenience. It is a requirement.
Autonomy level: Fully autonomous. Runs on schedule, posts results directly. No decision-making capability.
Role 2: Intelligence officer
Mission: Monitor information feeds, filter for relevance, and deliver concise briefings.
What it does:
- Monitors the research dump on a schedule
- Filters results based on predefined criteria (language pair, specialization, deadline proximity)
- Compiles a prioritized daily briefing delivered to a single channel
- Flags high-urgency opportunities separately from routine leads
What it does not do:
- Make decisions about which opportunities to pursue
- Contact clients or respond to postings
- Access any client-facing systems
Autonomy level: Fully autonomous within configured parameters. No human approval needed for monitoring, filtering, or briefing.
Role 3: Strategist
Mission: Researching, brainstorming, outlining, decision analysis, and strategic refining.
What it does:
- Collaborates with you to achieve clarity and positioning
- Analyzes opportunities and helps evaluate trade-offs
- Serves as a brainstorming partner for strategic decisions
What it does not do:
- Send communications on your behalf
- Make decisions
- Operate without your direct involvement
Autonomy level: Collaborative only. Every output requires your review and approval before it reaches the outside world.
You ask for an Eisenhower Matrix, your strategist complies. You ask for a cost-benefit analysis, it complies. You ask for a SWOT analysis, it complies. You ask for a JTBD evaluation, it complies. For refining content, the AIDA (Attention, Interest, Desire, Action) framework can help too. Other frameworks you might find useful are the Value Proposition Canvas and the USP (Unique Selling Proposition) Analysis.
The spine: A visual board
Purpose: Track tasks, deadlines, and workflows through a visual pipeline.
Recommended implementation: A Kanban-style board (such as Trello) with drag-and-drop cards moving across workflow stages.
Why visual matters: Solo professionals typically track work in spreadsheets, email threads, or memory. All three fail at scale. A visual board externalizes your project state — you see your entire pipeline at a glance instead of trying to reconstruct it mentally. When you can see three cards in ACTIVE and two in DELIVERED, you know your workload without having to think about it.
This is the board structure that works for me:
Board 1 — Translation Pipeline (primary revenue tracking)
| INBOX | TO BID | ACTIVE | DELIVERED | INVOICED |
| New leads from the intelligence officer | Bids in preparation | Work in progress | Awaiting payment | Paid and archived |
Board 2 — Content Calendar (blog and marketing)
| IDEAS | OUTLINED | DRAFTING | EDITING | PUBLISHED/DONE |
| Raw topics | Structure confirmed | Writing in progress | Review and polish | Live on site |
Board 3 — Business Development (client pipeline)
| PROSPECTS | REACHED OUT | IN CONVERSATION | PROPOSAL SENT | OUTCOME |
| Identified opportunities | Initial contact made | Active dialogue | Formal proposal delivered | Engagement confirmed or not |
Practical implementation tips:
- Install the browser extension for frictionless capture — when the intelligence officer surfaces a lead, clip it into INBOX without leaving your current tab
- Install the desktop application and mobile app for quick updates from any device
- Limit yourself to 3-4 boards. More than that creates its own management overhead
- Use 5-7 color labels maximum for priority and category (e.g., urgent, direct client, framework agreement, routine)
- Set up automation rules: when a deadline approaches, auto-label as urgent; when all checklist items are complete, auto-advance the card to the next column
Autonomy level: Fully autonomous for reminders and automated rules. Manual for card creation and movement ; these represent your decisions.
The autonomy framework
The most consequential design choice in this system is not which tools you select. It is where you draw the line between autonomous operation and human judgment.
Tier 1 — Fully autonomous (no approval needed)
- Data scraping and monitoring
- Daily briefing compilation
- Job lead filtering and prioritization
- Project board reminders and automation rules
- Scheduled reports
Tier 2 — Approval required (human reviews before action)
- Bid submissions
- Content publication
- Purchase decisions
- Any outbound communication
- Any action carrying financial or reputational consequence
Tier 3 — Human only (no delegation)
- Client relationships and negotiations
- Strategic direction and specialization decisions
- Creative judgment on translation quality, register, and the domestication-foreignization balance
- Ethical decisions about which work to accept
- Final editorial voice on all published content
This framework is not about trust in technology. It is about clarity on where your professional value resides — and designing a system that protects that value from dilution by operational noise.
What to expect: week one
This is not a universal template. It reflects one implementation trajectory. Your version will differ based on your specialization, client base, and working patterns.
| Day | What happens | What you’ll feel |
| Day 1-2 | Set up one board (Translation Pipeline), five columns, current work as cards. Configure the research assistant on an isolated machine. | Mild setup fatigue. Resist the urge to over-customize. |
| Day 3 | First automated scraping results arrive in your communication channels. Intelligence officer delivers first briefing. | Surprise at volume. Most of it irrelevant. That’s normal. Tighten your filters. |
| Day 4-5 | Filters improve. Briefing shrinks from 20 items to 5. You clip your first lead into the board from the browser extension. | The system starts to feel real. You’ll still check job boards manually out of habit. |
| Day 6-7 | First full cycle: lead arrives overnight, briefing in the morning, bid drafted after brainstorming and assessment with strategist’s powerful frameworks, card moves across the board. | The moment you realize you didn’t open a single job board today. The silence is strange. Then it’s not. |
After week one: Resist two temptations. The first is adding complexity (more boards, more labels, more automations). The second is abandoning the system after missing a day. Neither impulse is productive. Start with one board. Add one automation rule. Run it for a week before expanding.
A rough system that you use every day outperforms a perfect system that you abandoned on Tuesday.
Adapting this framework
This guide describes a system, not a product. The specific tools will change. The roles will remain.
If you implement nothing else, implement the separation: stop interleaving high-value craft with low-value scanning. Stop holding your project state in memory. Stop spending your mornings on work that a scheduled process could have completed overnight.
And if you handle confidential documents — legal, humanitarian, corporate, institutional — implement the isolation first. A dedicated machine with no access to client data, no connection to your primary accounts, no bidirectional data flow. That architecture is not optional for professionals in our field. It is the starting point.
Build the framework that matches your practice. Then do the work that only you can do.
This guide is part of the Words We Trust Resource Hub — working references for translation professionals navigating evolving practice.