Industries / Finance and business services / Staffing and recruiting
Finding candidates
Getting word of an opening to the people who might fill it, through public employment offices, job ads, job boards, search engines, referrals and recruiters who seek people out.
7
Milestones achieved, from 1909 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1909
Britain opens labor exchanges
Britain's Labour Exchanges Act of 1909 set up state offices where employers listed vacancies and job seekers registered. They were run by the Board of Trade through a new department directed by William Beveridge, and 430 opened in the first two years. In 1914 they registered 3,442,452 applications and filled 1,116,909 vacancies, by the count of a 1916 US Bureau of Labor Statistics bulletin.
Credit
Board of Trade; William Beveridge
1994
The Monster Board launches
Jeffrey Taylor, who ran a recruitment advertising firm, founded The Monster Board in April 1994, putting job ads on the web. Its later owner, Monster Worldwide, called it "the 454th commercial website in the world" in its 2004 annual report.
Credit
Jeffrey Taylor; TMP Worldwide, later Monster Worldwide
2015
Referrals measured at nine firms
Using personnel data from nine large firms in call centers, trucking and high tech, four economists found that applicants referred by employees were more likely to be hired and to accept offers, though their skills were similar. Referred workers were much less likely to quit, and in call centers and trucking they earned the firms more profit per worker.
Credit
Stephen Burks, Bo Cowgill, Mitchell Hoffman and Michael Housman (Quarterly Journal of Economics, 2015)
2017
Google gathers job listings
On 20 June 2017 Google began showing job listings gathered from across the web directly in its search results in the US, filterable by commute time and hours, with listings from partners including LinkedIn, Monster, CareerBuilder, Glassdoor and Facebook.
Credit
Google
2021
Regulator halts BOSS Zhipin sign-ups
BOSS Zhipin, a Chinese hiring app, lets job seekers and recruiters chat directly and swap résumés only when both agree; its owner Kanzhun says it was first in China to do this at scale. After the Cyberspace Administration of China announced a cybersecurity review on 5 July 2021, the app had to stop taking new users in China until 29 June 2022. By Kanzhun's count it then had 100.8 million verified job seekers.
LessonA hiring app that holds millions of personal profiles can be closed to new users for a year by a data regulator.
Credit
Kanzhun; Cyberspace Administration of China
2022
Weak ties tested on LinkedIn
Researchers analyzed randomized tests of LinkedIn's "People You May Know" feature, which changed how many weak ties, meaning distant acquaintances, it suggested to more than 20 million people over five years. In that time 2 billion new ties and 600,000 new jobs were created. Moderately weak ties led to the most job moves, and strong ties mattered more in less digital industries.
Credit
Karthik Rajkumar, Guillaume Saint-Jacques, Iavor Bojinov, Erik Brynjolfsson and Sinan Aral (Science, 2022)
2025
Monster and CareerBuilder bankrupt
Monster and CareerBuilder, merged in 2024, filed for Chapter 11 bankruptcy on 24 June 2025 and agreed to sell their job boards to JobGet, an app for gig and hourly workers. The chief executive blamed an uncertain economy, and the Associated Press report notes that job seekers had moved to Indeed, Glassdoor and LinkedIn.
LessonThe job boards that first took job ads online lost their audience once search engines and professional networks gathered postings in one place.
Credit
Monster and CareerBuilder
Watch for
Hiring Assistant time savings
LinkedIn says early users of Hiring Assistant, an AI agent that searches for candidates and helps write outreach, save more than 4 hours per role, review 62% fewer profiles and get 69% more acceptances of their InMail messages. It gives no sample size, time period or baseline.
Achieved whenAn outside study compares recruiters using the agent with a matched group who do not, and publishes time per role and reply rates.
Claimed by
LinkedIn, September 2025
Open
01
Ontario posting disclosure tally
A public, repeated count of how Ontario job postings meet two rules in force since 1 January 2026. Employers with 25 or more employees must state whether a posting is for an existing vacancy, and must disclose if they use AI to screen, assess or select applicants. Draw postings each month, record whether each statement is present and what it says, and publish the table.
The briefCredit · unclaimed02
Sourcing test with planted answers
A pool of 20,000 made-up professional profiles in which the right people for 100 search briefs are planted in advance. Some are easy to find (matching job titles) and some hard (the right skills under another title, or a recent career change). Score sourcing agents and plain keyword search on how many planted people appear in their first 50 results and how many wrong people they return.
The briefCredit · unclaimed
Open 01 · the brief
Ontario posting disclosure tally
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, repeated count of how Ontario job postings meet two rules in force since 1 January 2026. Employers with 25 or more employees must state whether a posting is for an existing vacancy, and must disclose if they use AI to screen, assess or select applicants. Draw postings each month, record whether each statement is present and what it says, and publish the table.
Why it is still open
Indeed Hiring Lab found that the share of Ontario postings on Indeed mentioning AI-related terms rose from 9% in October 2025 to 28% in May 2026. Its tracker flags words such as "artificial intelligence" anywhere in a posting, not the required statement, and it does not look at the vacancy statement or give a compliance rate. Estimates of "ghost jobs", ads with no real opening behind them, also exist; Hunter Ng (2024) estimated that up to 21% of Glassdoor ads may be ghost jobs. Those estimates infer intent, and none reads a statement the employer is now required to make.
Built with what exists
Canada's Job Bank open data, a monthly file of postings with occupation codes and posting dates, to draw the sample; a browser agent such as Browser Use to open each posting (see the Note "Deleting 80% to start over"); a language model that answers fixed yes-or-no questions about each posting (see the Note "A thousand questions, one pass"), with a person checking a sample; a public spreadsheet.
First step
Draw 200 Ontario postings from the October 2026 Job Bank file, open each one and code the two statements by hand to fix the coding rules.
Done looks like
At least 2,000 Ontario postings a quarter for four quarters, each coded for both statements, with employer size recorded where it is public, a person re-checking 10% of codes and the agreement rate published.
Open 02 · the brief
Sourcing test with planted answers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A pool of 20,000 made-up professional profiles in which the right people for 100 search briefs are planted in advance. Some are easy to find (matching job titles) and some hard (the right skills under another title, or a recent career change). Score sourcing agents and plain keyword search on how many planted people appear in their first 50 results and how many wrong people they return.
Why it is still open
The benchmarks found score agents on searches for real people, judged after the fact. xbench's recruitment set (June 2025) uses 50 real headhunting tasks, and an August 2026 paper on a sourcing agent judged 21 systems with an answer list built by several AI models, because a people search has "no acceptance criterion". Exa, a search company, publishes a People Search benchmark of 1,400 role queries that scores searches for real profiles on the open web, and its own system leads the table. LinkedIn's figures for its own agent come from LinkedIn alone. No test with a fixed, known answer list was found.
Built with what exists
Claude or another language model to write the profiles from a fixed recipe; a search index such as OpenSearch; open embedding models; agent frameworks that can search a supplied pool; GitHub or Hugging Face to publish.
First step
Write ten briefs, plant five right people for each in a pool of 1,000 profiles, and check whether keyword search misses the hard ones.
Done looks like
The pool, the 100 briefs and the answer lists published, with the share of planted people found in the top 50 and the share of wrong results reported for at least five systems, including a keyword baseline, and a person checking 200 profiles for realism and for planted answers that are wrong.