Lagou Is Dead: A 13-Year Job Board Killed by AI Pulling Out All Three Middleman Steps at Once
Lagou is dead. It topped the Chinese dev-community charts this week, and the dominant story in the comments is "internet winter," "BOSS Zhipin owns the market now," "the founder walked away." Those are outcomes, not causes.
For readers outside China: Lagou was, for most of the 2010s, the go-to job board for tech here — where engineers went to find work at internet companies, and where those companies went to find engineers.
The real cause sits one layer down. The three middleman actions Lagou lived on — writing the job description, screening resumes, running the first interview — were pulled one by one into corporate HR toolchains and candidate-side toolchains over the last 18 months. Lagou didn't lose to a competitor. It lost to its line of business being absorbed.
This is not an isolated recruiting story. It's the marker for AI's first replacement wave landing on middleman work, and there is a long queue behind it.
What Lagou actually sold for 13 years
When Lagou launched in 2013 the positioning was clear: vertical hiring for internet jobs, matching company JDs with engineer resumes.
But as a platform, its actual revenue came from three specific actions:
- Companies paid it to "package the JD" — turn the HR draft into something an engineer would want to click
- Companies paid for its resume screening — take 100 resumes down to 10 candidates
- Companies paid for outsourced first rounds and assessments — take 10 candidates down to 3 who go onsite
Those three things carried Lagou for 13 years. BOSS Zhipin never actually killed them; it just changed the shape of the matching.
What killed all three was AI.
How AI removed all three in 18 months
Step one: writing the JD (from H2 2024)
Starting with GPT-4, "AI JD assistants" appeared inside corporate HR teams. Early on it was a prompt template:
Write me a senior frontend engineer JD, stack is React + Next.js, 3-5 years experience
By the first half of 2025 every mainstream HR SaaS vendor here — Beisen, Moka, Kenexa — had folded JD generation into a one-click feature. HR departments stopped needing outside "packaging," and that line of platform business went from something to nothing.
Step two: resume screening (all through 2025)
Resume screening was the platform's biggest cash cow, and the easiest thing for AI to eat. The reason: a resume is highly structured text, and AI does this more accurately than the platform's human screeners.
| Dimension | Platform human screening | In-house AI screening |
|---|---|---|
| Cost per resume | ¥3-8 | ¥0.05 |
| Time per resume | 30 sec | 0.4 sec |
| Top-N recall* | 72% | 83% |
| Auditable for bias | Hard | Traceable in logs |
* Recall = of the candidates eventually hired, how many this screening step actually surfaced * These figures are estimates from interviews with 8 corporate HRBP / TA leads over the past six months, not published statistics — public data in this industry is scarce, so treat them as orders of magnitude
The kill shot is recall going the wrong way for the platform. The moment corporate HR notices that AI screens more accurately than the platform does and costs 60x less, that line of business has no reason to exist.
Step three: first-round interviews and assessments (late 2025 into early 2026)
The last step fell last, because a first interview is a conversation and the text loop is hardest to close there. Two things changed that in late 2025:
- Multimodal agents can run voice interviews, and the candidate experience beats "the HR screen you can never get scheduled"
- Model agreement on open-ended questions — system design, behavioral — reached 80%+
By Q1 2026 Lagou's customer payment rate fell off a cliff. All three steps were being handled by companies themselves, with AI. What remained for the platform was one function, "traffic entry point" — and a traffic entry point doesn't feed a 13-year-old company.
Not a winter. The middleman business got absorbed.
To see this clearly, swap "job board" for the more abstract term: middleman platform.
Middleman platform = takes unstructured demand from side A, processes it into structured input side B can use, charges for the processing
Job boards work that way. So do plenty of others:
| Middleman platform | The "processing" it sells | AI absorption |
|---|---|---|
| Job boards (Lagou) | JD writing / resume screening / first round | Done → shut down |
| Homework / ghostwriting platforms | Answer generation | 100% → a batch already died |
| Dev outsourcing platforms | Requirement breakdown / engineer matching | 70% |
| Legal consultation platforms | Case description / statute matching | 60% |
| Design marketplaces | Brief breakdown / style matching | 50% |
| Online 1:1 tutoring platforms | Student level assessment / course matching | 40% |
| Matchmaking / dating platforms | User profiling / matching | 30% |
| Real estate agencies | Requirement description / listing matching | 20% (heavy offline component) |
The moat of a middleman platform is the processing action. The speed at which AI absorbs that action into the toolchains upstream and downstream is the speed at which these platforms die.
* The absorption percentages in the table are subjective. I scored them on a gut read of "does a paid AI feature that replaces this processing step already exist in the toolchain upstream or downstream, and are people using it." Not market research — a directional call. Don't build a strategy off this table; do use it to see the queue.
If you work in a middleman business: 3 migration signals
If you're in a middleman platform or a middleman business — product, engineering, ops — don't wait for the "we have a three-year strategy" speech. Watch these three signals. All three lit means start moving.
Signal 1: your average deal size has fallen for 12 straight months There are only two reasons deal size drops in a middleman business: a competitor undercutting you (temporary), or AI replacement (permanent). The second one doesn't reverse.
Signal 2: you're adding an "AI assistant" to your core product The team sensed the threat, so it bolted on AI. But a middleman business adding an AI assistant is self-harm — the better it works, the more you've proven that AI can do this thing, and next contract the customer goes straight to the API and around you.
Signal 3: customers start asking for API access or on-prem deployment They want to run the AI themselves. Once customers start asking that question, the platform's date of death is already on the roadmap.
A sharper question for the veterans
Lagou is gone, and the question worth asking isn't "what happens to recruiting." It's "am I in a middleman role myself."
A lot of veterans haven't noticed that they are:
- PMO: processes requirements between business units and engineering teams
- The architect / system designer who doesn't write code: processes proposals between business and implementation
- The TL or manager without a concrete business line: processes information between levels
- Trainers and internal instructors: process content between the knowledge source and the students
What these roles share: income comes from a processing action, not from creating or deciding.
In the AI era, the marginal cost of a processing action trends to zero. Every role, company and platform that eats off one is standing in the same queue Lagou was in.
The question isn't whether your turn comes. It's what number you're holding.
Following on from 8-Line Spec Beats 8-Hour Prompts, which argued that AI outsourced "writing code" and made "describing the problem clearly" the scarce skill: this is the same mechanism at industry scale. AI outsourced the processing action, and creating / deciding became the scarce action.
The middleman is collapsing at both ends: the role layer and the platform layer.