1Tom tells it once
Tom's brief
Confirmed- 40,000 words · 3 linguists
- Due Fri 13 Nov, 5 pm
- Check-ins: 50% and 75%
- Flag if more than 15% behind
- Backup: Piotr
- Ask first
- Numbers only: no text
Tom tells the agent once: who has what, the deadline, and when to speak up.
2It asks at halfway, and again at three-quarters
- AskWords done so far?
- One-tap replya number
- Comparewith pace needed
- Flag or log15% rule
- Log it
No reply in 3 hours? Tom is called.
Twice before the deadline it asks each linguist one question: how many words are done?
Already track progress in your translation management system? You may not need this.
3One line falls behind
At halfway everyone is close. At three-quarters, one line has dropped below the limit.
4It asks. Tom chooses.
3 of 3 replies in
Lena is behind
Asking Piotr now: 3,500 words, due Fri 10 am. Status note for the client is drafted for you to send.
Piotr confirmed 2:30 pm ✓
Nothing changes until Tom picks. The agent drafts the messages either way.
5Delivered
Status: all parts
- Mara · 15,000 wordsDelivered Thu 12 Nov, 3 pm
- Jonas · 15,000 wordsDelivered Thu 12 Nov, 5 pm
- Lena 6,500 + Piotr 3,500Delivered Fri 13 Nov, 10 and 11 am
- Deadline: Fri 13 Nov, 5 pm
- Check-ins stopped
Tom found out on Wednesday, not on Friday. The agent stops asking.
Built-in limits
- Never sees the textOnly word counts and dates.
- Asks firstNothing moves without Tom's yes.
- A number, not a fileLinguists reply with one figure.
- Logs everythingEvery question, reply and choice.
Same pattern, your business
-
Home improvement example page
Job is halfway through its days but under half the tasks are tickedthenyou choose: add a helper, move the end date, or wait
-
Short-term rentals example page
Cleaner hasn't confirmed by 1 pm on turnover daythenjob offered to your backup cleaner
-
Translation agencies This page
Nobody accepts a job in timethenoffered to the next linguist
-
Editing & small presses example page
Edit is three-quarters through its time with under 60% of chapters backthenyou choose: extend, add a second editor, or wait
Illustrations, not client results.
Which deadline would you like an agent to watch?
Book a free discovery callFree 30-minute discovery call · Plain-English recommendations · No obligation
More details
Tom's full brief
Tom is a project manager at a small agency (made up). A 40,000-word legal job is split across three linguists with a Friday 13 November, 5 pm deadline.
| Setting | Tom's choice |
|---|---|
| Job size | 40,000 words, legal |
| Parts | Mara 15,000 words, Jonas 15,000 words, Lena 10,000 words |
| Window | Mon 2 Nov to Fri 13 Nov, 10 working days |
| Check-in marks | 50% = Mon 9 Nov, 9 am; 75% = Wed 11 Nov, 1 pm |
| Words needed at each mark | Mara and Jonas: 7,500 / 11,250 / 15,000. Lena: 5,000 / 7,500 / 10,000 |
| Behind limit | 15% |
| No-reply rule | No reply after 3 hours: Tom is called |
| Backup | Piotr |
| Approval | Ask first |
| What the agent sees | Numbers only: names, dates and word counts. Never the text |
The ten chat bubbles in the first scene are the plain-language way of filling this in:
| Who | What was said |
|---|---|
| Agent | What's the job? |
| Tom | 40,000 words, legal. Three linguists. |
| Agent | Who has how much? |
| Tom | Mara 15,000. Jonas 15,000. Lena 10,000. |
| Agent | Deadline? |
| Tom | Fri 13 Nov, 5 pm. |
| Agent | When should I check in, and how far behind is too far? |
| Tom | Halfway and three-quarters of the way. Tell me if anyone is more than 15% behind. |
| Agent | Do you want me to read the text? |
| Tom | No. Numbers only. Ask me before you change anything. |
What happens at each check-in
- At the mark, it sends each linguist the same one-line question: "Roughly how many words are done so far? Reply with a number."
- It collects the replies (one tap or one number). A reminder goes out after an hour; this is a setting.
- It works out the words that should be done by now: share of the part times the percentage of time used.
- It works out how far behind or ahead each person is: (needed minus done) divided by needed.
- Anyone more than 15% behind is flagged, and the agent messages Tom with the options.
- Anyone who hasn't replied after 3 hours: Tom is called, whatever the number.
- It logs the question, the reply and the result.
The arithmetic is simple. The agent does not predict who will be late. It compares a number with a line.
The full numbers behind the chart
All numbers are made up. This table is also the data behind the burn-up chart.
| Linguist | Check-in | Words done | Words needed | Share of part done | Behind or ahead | Flagged? |
|---|---|---|---|---|---|---|
| Mara | 50%, Mon 9 Nov | 7,700 | 7,500 | 51% | 3% ahead | No |
| Mara | 75%, Wed 11 Nov | 11,500 | 11,250 | 77% | 2% ahead | No |
| Jonas | 50%, Mon 9 Nov | 7,300 | 7,500 | 49% | 3% behind | No |
| Jonas | 75%, Wed 11 Nov | 11,100 | 11,250 | 74% | 1% behind | No |
| Lena | 50%, Mon 9 Nov | 4,600 | 5,000 | 46% | 8% behind | No (inside the 15% limit) |
| Lena | 75%, Wed 11 Nov | 5,700 | 7,500 | 57% | 24% behind | Yes |
Team total at 75%: 28,300 words done of 30,000 needed, about 6% behind. A team total can look fine while one part is late. That is why the agent checks each person against their own share.
After Tom chose: Piotr took 3,500 words (confirmed Wed 2:30 pm, delivered Fri 10 am). Lena kept 6,500 words and delivered Fri 11 am. Mara delivered Thu 12 Nov, 3 pm, and Jonas Thu 12 Nov, 5 pm.
What if...
- Lena replies that she's fine, but the number is behind. The agent shows the number and lets Tom decide.
- Nobody replies in 3 hours. Tom is called.
- A linguist is ahead. It's logged, with no message.
- Tom picks "Wait". The agent asks again at a time he sets, or at the next mark.
- Tom picks "Ask the client for more time". The agent drafts the message, and Tom sends it.
A client's contract may restrict who can touch a job. The agent only offers work to linguists Tom has already approved.
Why the text stays private
The agent works from numbers, names, dates and the job's metadata. It never opens the source or translated files, never stores them, and never includes them in a message.
Files go from Tom (or his own system) to the linguist exactly as they do today. The agent's message to a backup linguist contains the word count, the language pair, the deadline and nothing else.
If a client's contract limits who may handle a job, or limits any use of AI, the agent can be switched off for that job.
Everything is logged, so Tom can show a client what was and wasn't done. See how we handle your data.
When you might not need this
If your translation management system already shows progress and sends deadline alerts, and your linguists update it, you may already have most of this.
The part that is usually missing is the habit: someone has to ask, compare, and decide at the right moment. That is what this agent adds. It asks people who don't use your system, it works out the pace line for you, and it brings you the options with the messages drafted.
If you have a small team and one or two jobs at a time, a calendar reminder and a quick email may be enough. The pattern is worth building when several jobs run in parallel and checking each one by hand keeps getting squeezed out.
It relies on linguists answering a quick question and on your agreements with them allowing it.
The business examples in full
These are illustrations too. The pattern is the same each time: set a pace, check in, compare, and keep a person in charge.
- Home improvement: a job is halfway through its days but under half the tasks are ticked. The agent asks you to choose: add a helper, move the end date, or wait.
- Short-term rentals: a cleaner hasn't confirmed by 1 pm on turnover day, so the job is offered to your backup cleaner.
- Translation agencies: nobody accepts a job in time, so it is offered to the next linguist.
- Editing and small presses: an edit is three-quarters through its time with under 60% of chapters back. The agent asks you to choose: extend, add a second editor, or wait.
How we'd start with you
We'd start with one workflow and a measurable goal, not a big project.
- Discover: a free call to understand your business, goals and where time is going.
- Assess: we map the workflow, measure the baseline and give you a prioritised AI roadmap with a fixed-price quote.
- Pilot: we build the first agent, demo progress weekly and launch it with human approval switched on.
- Scale & run: once the results are proven, we add agents, form teams and keep everything monitored and improving.
We'd start with one workflow, one team and approval switched on.