LinkedIn's Hiring Assistant Hit $450M ARR. It Still Can't See GitHub.

If your last five engineering hires came from a GitHub thread, a Discord, or a paper citation, the $450M question is whether LinkedIn's new sourcing agent can find the sixth. It cannot, and the reason is structural, not temporary.

By April 2026 LinkedIn's Hiring Assistant was running at a roughly $450M annualized rate, the first time the company has ever broken out revenue for one of its AI tools. That number is real, and it is also the exact reason the coverage gap is about to get worse.

What LinkedIn Hiring Assistant is, and why $450M matters

LinkedIn Hiring Assistant is an agentic sourcing product that reads a job description, generates a candidate list, drafts outreach, and manages replies inside LinkedIn Recruiter. It hit approximately $450M in annualized revenue by April 2026, per The Information, growing from a 500-company charter in late 2024 to 8,000+ early users by September 2025.

The product is working on its own terms. LinkedIn's own charter data shows customers:

• Reviewed 62% fewer profiles per role
• Saved 4+ hours per role
• Saw 69% higher InMail acceptance

"Recruiters told us half their day was low-value work, so we made a bet on understanding their pain to get our solution right," Dan Shapero, LinkedIn's new CEO, told Reuters. February 2026's quarterly drop added AI Applicant Targeting, AI Follow-Ups, Verified Applicant Spotlight, and Microsoft Teams collaboration. Charter customers include AMD, Canva, Siemens, and Zurich Insurance.

None of that is the argument against it. The argument is that every dollar of that $450M is spent inside a walled garden LinkedIn is actively litigating to keep closed, and the agent inherits the walls by construction.

$450M

Why "62% fewer profiles reviewed" is a coverage problem, not a productivity win

An agent that reviews 62% fewer profiles is not searching more of the market. It is searching less of it, harder. That is fine when the underlying index already contains everyone you want to reach. It is a compounding problem when it does not.

The mechanism is base-rate bias. If LinkedIn's index is missing or mis-titling a candidate, a human recruiter with a Boolean string might still stumble into them through a second-degree connection or a keyword scan. An agent that pre-filters aggressively will not. The candidate goes from "hard to find" to structurally invisible, because the shortlist never had them to prune in the first place.

This matters most in exactly the segments where hiring is competitive:

• Deep infrastructure roles where the signal lives in GitHub commits and RFC threads, not LinkedIn skill endorsements
• Fast-relabeling titles (AI Engineer, Forward Deployed Engineer) where LinkedIn profiles trail reality by 6-18 months
• Thin geographic pools where missing 10 people means missing the role

The Rust-in-Germany case: 188 people, and a 10% recall miss ends the search

In Refolk's index, only 188 current Software, Senior, or Staff Engineers in Germany list Rust as a skill. In the US, the equivalent Rust, Go, or Kubernetes pool is 29,559. When your market is 188 people, coverage is not a nice-to-have. A 10% recall miss is 19 people, and 19 people is often the entire viable shortlist.

Here is what the numbers look like side by side, drawn from Refolk's index of professional profiles and the public sources cited above:

The 188 Rust engineers in Germany are concentrated in Berlin (5), Munich (4), and Düsseldorf (2), sitting at Google, Bosch, SAP, Helsing, and DeepL. A meaningful share of the interesting ones are Rust maintainers whose LinkedIn headline still says "Software Engineer" and whose actual Rust identity lives in a GitHub bio and a couple of crates. A LinkedIn-only agent, no matter how good the model, cannot see that.

This is the exact gap Refolk closes: you describe the person in plain English ("Rust maintainer in Germany with kernel or systems work"), and Refolk pulls candidates across GitHub, LinkedIn, and the open web into a single ranked shortlist, keyed on the code they actually shipped, not the skills they remembered to tag.

The "AI Engineer" relabel is happening faster than LinkedIn titles refresh

LinkedIn's structured search punishes title drift, and title drift is currently the fastest-moving variable in AI hiring. In a Refolk sample of ML/AI-titled US roles, "AI Engineer" already runs at roughly 56% of "Machine Learning Engineer" volume (9 vs 16 of 25). Six months ago that ratio was noise. Today it is a category.

The candidates driving that shift are the ones you want most. They are the people whose GitHub README says "AI Engineer building agent evals," whose personal site says the same thing, and whose LinkedIn profile still says "Senior ML Engineer" because they last edited it in 2023. A Boolean filter on title:"AI Engineer" inside LinkedIn Recruiter misses them. An agent built on top of that Boolean filter misses them faster and more confidently.

The general shape of the problem:

1. Title conventions in AI shift on a 6-month cycle
2. LinkedIn profiles refresh on a 12-24 month cycle
3. GitHub bios, HuggingFace pages, and personal sites refresh on a 1-3 month cycle
4. An agent locked to source 1, filtered against source 2, will lag source 3 by 6-18 months

An agent that reviews 62 percent fewer profiles is not searching more of the market. It is searching less of it, harder.

LinkedIn's lawsuits are the product roadmap

If you want to know what LinkedIn's agent will and will not do in 2026, read the docket, not the changelog. LinkedIn spent the same twelve months it shipped Hiring Assistant suing the two largest third-party pipes into its data out of existence.

• January 2025: LinkedIn sued Proxycurl (Nubela), a scraping API doing roughly $10M in annual revenue that thousands of recruiting and sales tools depended on. By July 2025, Proxycurl had settled under a permanent injunction, deleted its LinkedIn data, and shut down entirely.
• October 2025: LinkedIn sued ProAPIs, alleging over one million fake accounts powering a scraping API sold for up to $15,000 per month. The parties reached an agreement in principle in February 2026.

This is not incidental. The moat under $450M of ARR is data exclusivity. Any "LinkedIn plus open web" pitch coming from LinkedIn itself is structurally unlikely, because opening the graph destroys the pricing power. That is why the multi-source category, GitHub plus LinkedIn plus the open web, is where third-party AI sourcing agents live and where LinkedIn Recruiter alternatives are being priced.

The verified-skills program is LinkedIn conceding the point

In late 2025 LinkedIn launched a verified AI skills program with Descript, Replit, Relay.app, and Lovable, with GitHub, Zapier, and Gamma named as partners joining "in the coming months." Instead of one-time exams, users are assessed on real usage inside those tools. This is LinkedIn conceding that self-reported profiles are no longer a sufficient signal for AI-era roles.

But importing verified badges is not the same as indexing the work. A badge that says a user spent time inside Replit is not the same signal as reading the actual repos, commit history, and code review comments. The code still lives off-platform. GitHub, notably, is on the "coming months" list, not the launch list, and even when it lands the integration is a badge, not an index.

Multi-source candidate sourcing solves this the other way around. Start with the work (GitHub commits, HuggingFace models, arXiv authorship, personal sites), then join to LinkedIn for tenure and title context. That is the order of operations agentic sourcing needs, and it is not an order LinkedIn's own product is allowed to run in.

The economics: 91% of switchers cut LinkedIn Recruiter spend

Per Pin's 2026 user survey, 91% of teams reduced or eliminated LinkedIn Recruiter spend after switching. That is not a marketing line, it is a purchasing pattern.

The cost gap explains the switch. UK G-Cloud pricing shows LinkedIn Recruiter seats at £6,750-£8,925 per year, with Hiring Assistant add-on tiers stacking another £13,500-£67,500 on top. Pin's top-tier Business plan at $249/month costs less than 40% of a single Recruiter Corporate seat annually. When the underlying index has the same structural gaps at either price point, the cheaper stack wins on math alone.

91%

The other number worth staring at: 53.8% of candidates now use niche platforms over LinkedIn alone, up from 49.2% the year before, per iHire's 2025 State of Online Recruiting report. Combined with InMail response rates stuck at 18-25%, the LinkedIn-only funnel is getting thinner from both ends: candidates are spending less time there, and the ones who are there are answering less often.

What a multi-source AI sourcing agent surfaces that LinkedIn's cannot

A multi-source agent surfaces three specific candidate categories a LinkedIn-only agent cannot see, no matter how good the model on top gets:

• The GitHub-native maintainer with a stale LinkedIn. Ships in Rust or CUDA, tags themselves accurately in a GitHub bio, has not edited LinkedIn in two years. Invisible to any Boolean or agent that filters on LinkedIn skills.
• The relabel-early AI engineer. Job title drifted from "ML Engineer" to "AI Engineer" on their personal site and GitHub in Q4 2025. LinkedIn still says ML Engineer. Structured-title agents miss them until they refresh.
• The open-web specialist. HuggingFace model author, arXiv co-author on a paper that matters to your product, blog with three technical posts. On LinkedIn they are one line: "Research Engineer at $BIGCO." The differentiating signal is entirely off-platform.

None of these people are exotic. In the US Rust, Go, or Kubernetes pool of 29,559, a meaningful fraction sits at Meta, Microsoft, Datadog, OpenAI, and Rubrik with LinkedIn profiles that describe a fraction of what they actually do. That is the population Refolk is built to find: you ask in plain English, and the ranked output reflects the code and the profile, not just the profile.

The takeaway for founders and heads of talent

LinkedIn's $450M ARR proves the demand for AI sourcing agents is real and durable. It does not prove that a LinkedIn-only agent is sufficient for GitHub recruiting, deep infrastructure hiring, or any market where the differentiating signal lives outside LinkedIn's graph. The lawsuits, the verified-skills program, and the 62% fewer profiles metric all point the same way: LinkedIn is optimizing depth inside its walls at the exact moment the interesting candidates are spending more time outside them.

Buy Hiring Assistant if your pipeline is already 90% LinkedIn-sourced and you want to do less clicking. If your last five hires came from a GitHub thread, a Discord, or a paper citation, buy the agent that can see those places too.

FAQ

Is LinkedIn Hiring Assistant worth the demand it is seeing?

Yes, in the sense that recruiters are clearly paying for agentic workflows and getting real time back (4+ hours per role, 62% fewer profiles reviewed per LinkedIn's charter data). The productivity gains are legitimate. The question is not whether the agent works, it is whether the index it works on covers your specific hiring problem. For LinkedIn-native funnels it does. For GitHub-heavy, AI-native, or thin-geography roles, it inherits the gaps of the underlying graph.

What can a multi-source AI sourcing agent find that LinkedIn's cannot?

Three categories: GitHub-native maintainers with stale LinkedIn profiles, engineers who have relabeled themselves as "AI Engineer" on GitHub or personal sites while LinkedIn still shows an older title, and open-web specialists whose signal lives in HuggingFace, arXiv, or blogs. Refolk's index shows the AI Engineer title already running at ~56% of ML Engineer volume in a US sample, so the relabel category alone is substantial.

Why does the Proxycurl shutdown matter for buyers?

It signals that LinkedIn will not tolerate third parties enriching or re-indexing its data at scale, and that any "LinkedIn plus X" product coming from LinkedIn itself is unlikely because it would erode the moat. If you need LinkedIn coverage combined with GitHub and open-web coverage, you need a vendor whose index was built to include all three sources natively, not one hoping LinkedIn will open up.

How do I evaluate LinkedIn Recruiter alternatives without running a six-month pilot?

Give the shortlist tool three specific queries where you already know the right answer: a thin geographic pool (like Rust engineers in Berlin), a fast-relabeling title (AI Engineer with agent-eval work), and a role where the winning candidate came from GitHub or a Discord, not LinkedIn. If the tool surfaces the person you actually hired inside the top 20 results, the index covers your problem. If it does not, no amount of agent polish will fix it downstream.