iampro.iolab
iampro.io lab · Research · Horizon 2027 to 2030
The iampro.io lab studies and builds what AI needs to look for work on people’s behalf, reliably.
AI assistants such as ChatGPT or Claude already help people look for work. Tomorrow they will act on their behalf: they will be agents. Postings, résumés, interviews and intermediaries will not go away: we observe how these AIs cope in a market that was not built for them, and we build what they are missing.
Five lines of work
Each one starts from a difficulty observed when a machine searches, explains or acts on someone’s behalf.
Check that a posting is open, reachable from the stated country, and posted by the employer or by an intermediary.
Back what an agent claims about a person without exposing their data or taking the control of their evidence away from them.
Carry a total, its counting rule, the sources and the context needed to read an answer correctly.
Show what a person still has to decide, verify and correct when the agent does part of the search.
Measure where agents add something, and where meeting in person, a human intermediary or local knowledge remain necessary.
Work
Two kinds of dated entries. A measure says what breaks when an AI works in the real market. An experiment says what we built in response, in service with early users.
107 on site 91 others, mostly limited to one country 2 open worldwide
In a random draw of 200 postings labelled remote, 107 actually required on-site presence, and only 2 were open from any country.
Each posting is read by a language model answering two separate questions: is the remote work real, and from which countries can the job be held. A second, larger model rereads the same sample. Scope, limits and data
27 postings kept by the assistant 81 never seen
Asked "who is hiring forklift operators in Dallas?", an assistant ran two searches, kept 27 postings and summed up the rest as "probably more". For the same area and freshness, 108 postings existed, from 82 employers.
The question was asked by the lab itself, to an assistant connected to a test account, while shooting a film. Direct count on the iampro.io index with the same area and freshness rule as the assistant's two searches, each of which returned at most 20 postings. Scope, limits and data
A person's own assistant, Claude or ChatGPT, works inside their iampro.io account: profile, evidence, search, applications, follow-ups. It is the ground of every measure the lab makes.
A person reviews what the machine proposes to a registered candidate and sends it in iampro.io's name. The human stays in the loop where judgment matters.
Every posting labelled remote is read: real or not, and from which country it can be held. Search drops what is known to be on site or reserved for a country other than the person's. Answer to measure 01.
Search returns to the assistant the total number of matching postings and employers, with the counting rule stated plainly, on top of the page shown. Answer to measure 02.
Recounted every hour: postings labelled remote, those read as on site, those not read yet. The first index the lab will publish.
Every new entry, measure or experiment, is added here with its date. Feed: RSS.
Where the ground comes from
iampro.io is a job search service open in every country. People connect it to their own assistant, Claude or ChatGPT among others.
That position makes it possible to observe what an agent does when it looks for work for someone: how it phrases a request, selects postings, sums up what it found, and sometimes hides what is missing.
Employers do not pay iampro.io to publish their postings. The lab can therefore examine postings as objects to verify, with no presentation to protect.
The lab itself reads the market with language models, and publishes the method and the margin that come with every measure.
Working rules
A date, so that an old state is never presented as a permanent truth.
A scope, to say precisely what was observed and what was not.
A method and its limits, so that the result can be discussed or reproduced.
A source with every piece of information that guides a decision.
No measure drawn from the activity of people registered on iampro.io. The lab’s work is about postings, employers, tools and their answers.
