DuckBill Gets 1,000 Applications a Day. Two Are Real. Source the Rest.
Pragmatic Engineer's July 8 survey of 50+ hiring managers confirms inbound recruiting collapsed in H1 2026. Here's the outbound playbook per profile.
Gergely Orosz published Part 3 of the 2026 tech jobs market series on July 8, based on interviews with 50+ hiring managers. The headline number belongs to Mike Julian, CEO of DuckBill Group: about 1,000 inbound applications a day, and maybe two worth looking at. If you still run a pipeline that assumes inbound is where the good candidates live, you are staffing off 0.2% signal against 99.8% AI slop.
This piece is a playbook for the hiring managers and recruiters Gergely surveyed. The channel didn't slow down. It inverted. Reading inbound now costs more than sourcing outbound, and the profiles those 50+ managers actually want (distributed systems specialists, product engineers with taste) are not in the pile anyway.
The math: 1,000 in, 2 out
Julian's ratio is the cleanest data point in the Pragmatic Engineer piece. One thousand applications per day, roughly two worth reading. That is 0.2% conversion at the very top of the funnel, before a recruiter screen, before a tech screen, before anything.
Cold outbound to a well-targeted engineer converts at 2 to 5% reply rate on a decent message. Do the arithmetic. Even at the low end, outbound is ten times more efficient than reading inbound. And the outbound candidate is, by construction, someone you actually want. The inbound candidate is a coin flip on whether they wrote their own resume.
James McWalter at Paces reported 23,000 applications in 30 days for 8 in-person New York roles in the prior installment. Nearly 3,000 per job. A robotics engineering manager in Canada told Gergely most applicants were "totally unqualified, and the qualified ones do poorly in an interview." An Engineering Director in the survey called inbound "a useless channel." This is not a Julian-specific problem. It is the state of the channel.
Why inbound broke: AI slop plus fake candidates
Two things converged. First, applying got free. Robert Walters reported that 65% of professionals in the Middle East region now use AI chatbots or CV builders to apply to dozens of roles at once, and 70% of employers there saw a surge in AI-driven applications. The same dynamic is running in the US and EU. LLMs will spray a tailored cover letter at every posting on LinkedIn in an afternoon.
Second, the fakes got good. A year ago the "AI faker" was a security curiosity, mostly framed around suspected North Korean state agents. Gergely's July 8 sources now describe it as routine: candidates using AI to generate resumes, generate portfolios, and coach through live interviews. If you hire remote, you are running an authentication problem before you run a hiring problem.
The result is a signal-to-noise collapse. LinkedIn Jobs has become, in one Pragmatic Engineer quote, "an irrecoverable hellscape for inbound applications." That same company kept LinkedIn Recruiter and turned off inbound entirely. Same platform, opposite outcomes. It is exactly what happened to Stack Overflow Jobs. Any pure inbound channel gets destroyed by AI; any tool that supports curated outbound wins.
Reading inbound is now more expensive than sourcing outbound. The channel didn't die. It inverted.
The catch-22 Gergely named
Here is the framing that makes the July 8 piece worth rereading: hiring managers say senior-plus engineers are "not available to be recruited, at the same time as experienced, proven professionals find their applications ignored by employers." A staff engineer in the survey put it cleanly: "It feels like everyone who has a good job is holding onto it for dear life, and THOSE are the people we want to hire."
Both halves are true. Q1 2026 saw 52,050 announced tech layoffs. AI-fluent seniors still fill in 17 days at $206K base. The market is bimodal. For AI Engineering, ML, and FDE, it is incredible. For everyone else, less so. So the person you want is employed, not scrolling job boards, and buried under recruiter InMails from twenty other companies who also just realized inbound is dead.
The way out is not to post more jobs. It is to name the specific engineer you want and go get them.
The two profiles nobody can find
Gergely's 50+ interviews converge on two profiles that keep coming up empty.
Distributed systems engineers
A hyperscaler recruiter in the piece has 20 open reqs for distributed systems engineers and no qualified pipeline. This is not a comp problem. It is a supply problem. Every CS grad from 2024 through 2026 optimized for the AI/ML track. Meanwhile, infra and distributed systems takes 5 to 7 years to develop and cannot be shortcut with Cursor. The pipeline drained just as hyperscaler capex exploded.
A tight search of professional-network data for senior and director-level distributed systems specialists in the US surfaces roughly 7,600 people total. Most of them cluster at Databricks, Starburst, Dragos, and ClosedLoop, in San Francisco, Seattle, Boston, and Austin. That is your addressable market. Not the 1,000-a-day pile.
If your job posting for a "Senior Distributed Systems Engineer" is generating 400 applicants a week, none of them are in that 7,600. This is the exact friction we built Refolk for: you describe the target in plain English ("senior distributed systems engineers who shipped consensus or storage work at Databricks, Starburst, Dragos, or ClosedLoop, open to Bay Area or remote"), and you get a ranked shortlist across GitHub, LinkedIn, and the open web. No boolean, no scraping, no reading slop.
Product engineers with taste
The second profile is harder. A tech lead at a seed-stage LA startup told Gergely: "'Product engineer' has been a hard profile to find. It is also hard to find someone with a decent design eye who can also build full stack. The hardest thing to hire for has been taste + trust. I'd rather hire someone who is 'behind' on AI, but has great taste and judgment than someone with complex agent setups and prompt libraries."
That is a big shift from the 2024 conventional wisdom, which was to hire for AI tooling depth. Now taste is a first-class criterion. And taste does not show up in a resume. It shows up in shipped side projects, in interface decisions, in what someone chose not to build.
LinkedIn does not help you here. The UK has 160K Software Engineers listed and only 1.4K Product Engineers, and the ones who exist command £100K to £150K-plus starting in London. The profile is thin because the label is new. What you actually want is a full-stack engineer with a portfolio a designer would respect, and there is no filter for that.
Palantir's Forward Deployed Engineer role is the closest existing archetype. FDEs ship product, sit with the customer, and make judgment calls. Sourcing product engineers looks a lot like sourcing FDEs: look at GitHub for taste, look at personal sites for design sensibility, look at Show HN for shipped work. Refolk lets you write that query directly ("full-stack engineers with strong design sense, shipped side projects, active on GitHub in the last 12 months, based in NYC or remote friendly") and returns candidates with the evidence attached.
The playbook: outbound by profile
The July 8 piece implies a channel strategy without stating one. Here it is explicit.
Turn off inbound where it's noise
For senior-plus roles at companies with any brand awareness, inbound is a tax. Turn off the LinkedIn Jobs posting or hide it. Keep the careers page for compliance and employer branding. Don't ask a recruiter to read 1,000 applications a day for two hits. That work has negative EV.
Source from signal, not from lists
Every profile in Gergely's survey has a signal that is not on a resume. Distributed systems engineers show up in the commit history of Kafka, Kubernetes, TiDB, or their employer's public repos. Product engineers show up in Show HN, Dribbble-adjacent portfolios, and personal sites that render well on a phone. FDEs show up in customer-facing case studies. Source from where the signal lives.
This is the second place Refolk earns its keep. Ask for "engineers who have contributed to distributed database internals in the last year" and it will actually cross-reference commit history, employer, and location, instead of returning a keyword search on the string "distributed systems." That is the difference between a shortlist and a wishlist.
Screen for taste with artifacts, not Leetcode
The LA tech lead's quote is the most important shift in the July 8 piece. If taste and judgment beat AI tooling depth, your screen has to test taste. Portfolios, product decisions, and shipped side projects. Ask the candidate to critique a UI. Ask them what they cut from their last project and why. Shraddha Sunil at MeetGinger is building interview-screening software specifically because AI has broken the standard signals; expect more of this category.
Verify identity early
If you hire remote, do live video the first call. Do not wait until the onsite. The fake-candidate problem is now common enough at US, UK, and EU remote-friendly employers that this is a table-stakes step, not paranoia.
Watch the second-order risk
One thing the July 8 piece does not flag: when inbound dies, hiring pipelines go dark. Companies fall back on personal networks, which are demographically narrow. If you are going all-outbound, be deliberate about widening the sourcing net. Tools that search the open web (GitHub, personal sites, conference talks) surface people your network does not know. That is a feature, not just for fairness but for competitive advantage. The best distributed systems engineer in Austin is probably not in your LinkedIn 2nd degree.
What Bay Area time-to-hire is telling you
Bay Area median time-to-hire for senior engineers went from 38 days in Q3 2025 to 67 days in Q1 2026. That is a 76% increase in six months. Every extra day is a day the AI-fluent senior your competitor also wants is closing their other offer.
Time-to-hire stretches when your funnel is broken at the top. It does not stretch because interviews got harder. It stretches because you spent three weeks reading 1,000-a-day inbound before admitting nothing in there was going to work, then started outbound cold. Skip step one. Start with the shortlist.
FAQ
Is inbound recruiting really dead in 2026, or just noisy?
For senior-plus roles at any company with brand awareness, it is functionally dead. Julian's 0.2% conversion at DuckBill, Paces' 23,000 in 30 days, and an Engineering Director in Gergely's survey calling inbound "a useless channel" all point the same direction. Inbound may still work for junior roles at unknown companies, but for the profiles Pragmatic Engineer's July 8 hiring market piece flagged, active outbound is the only viable channel.
Why is the distributed systems engineer shortage getting worse?
Two structural reasons. CS grads from 2024 to 2026 optimized for AI/ML because that is where the hype and the money were. Distributed systems mastery takes 5 to 7 years and cannot be shortcut with LLM tools. Meanwhile, hyperscaler capex exploded and every AI company needs infra. The pipeline drained just as demand spiked. Expect the shortage to persist through 2028.
How do you source "product engineers with taste"?
Not through LinkedIn Jobs. Look at Show HN, personal portfolio sites, GitHub for shipped side projects, and Dribbble-adjacent work for design sensibility. The Palantir FDE archetype is the closest template. Screen with artifacts (portfolios, product critiques, past decisions) rather than Leetcode. The LA tech lead in Gergely's survey was explicit: taste and judgment beat AI tooling depth right now.
What is the fastest way to replace an inbound funnel?
Pick one target profile, name it precisely (employer clusters, geography, evidence of the work), and generate a ranked shortlist from GitHub, LinkedIn, and the open web. Reach out with a message that references the specific work, not the job description. That takes an afternoon and typically converts at 2 to 5%, which beats reading a month of inbound. Refolk was built for that first step; the rest is your outbound message and your interview loop.