Ending the AI Arms Race in Recruitment
- olivermorris83
- 52 minutes ago
- 6 min read
When cars first appeared they naturally used the 19th century lanes threaded between villages. The results were predictable: blind bends, horrendous accidents and regulation demanding flag-waving “safety runners”. Not until roads were rebuilt for motor-traffic did driving become safe, fast, and routine.
Nowadays AI is the new technology and traditional recruitment processes are the lanes overwhelmed with traffic. Both recruiters and candidates are targeting the other with AI firepower. Recruiters are swamped with irrelevant AI generated resumes and candidates are either ghosted or subject to the trials of Hercules. It's an antagonistic market.
Each new plug-in—another résumé parser, another cheat extension—merely aggravates the arms race, the human participants on both sides are caught in the cross fire. What we need is a process designed to realise the technological benefits for both.
Fresh Hope
I have recently had a glimpse of what the solution may be and it comes from a modest source. Two new standards have been agreed and adopted by the hyper scalers which allow AI agents to talk with other agents. It may sound innocuos, but this entirely changes the game's dynamics. Co-operation may yet be the best strategy, not adversity. The standards are:
Model Context Protocol (MCP)
a secure way for AI agents to pull or push data to/from business tools (Home)
Agent-to-Agent Protocol (A2A)
a vendor-neutral standard allowing AI agents to negotiate and co-operate
Think of MCP as on-ramps to data warehouses, and A2A as the highway code that lets thousands of agents merge safely at speed.

Meet the Lean Triad
With those rules in place we can redesign hiring around just three specialised agents. This is only a proposal, objections and suggestions are welcome. But I see startups are already within arms reach of realising the potential.
Picture the new cast of AI agents: Candy works for candidates, a dedicated side-kick who learns their ambitions and ability; Busby is a business's HR agent who understands the employer and their requirements deeply; and Marco is the market maker, matching participants such that candidates and businesses know the process works in their interests and learn to trust it.
Agent | Works for | Core duties |
Candy | Candidate | Intelligently converses with candidates to captures what they are really seeking, what they will compromise on, and they won't. It learns both the deal breakers and the changes they'd move heaven and earth to make. Candidates trust it with psychometrics and other facts they don't want on LinkedIn, or shared with employers. Of course, it also has access to a resume. |
Marco | Marketplace matcher, neutral and audited | Runs a stable-matching algorithm (Gale-Shapley) to pair roles with candidates; opens a secure channel for each match; logs and audits every step (Harvard EECS). |
Busby | Business/employer | Employed by the organisation, it learns over many hiring episodes, knows the policies and departments involved. It interviews the requisition creator, saving them time to turn fuzzy needs into a structured role, checks comp bands & compliance, puts final offers in context, and feeds onboarding data back to HRIS. |
How it Works
Line managers no longer email ad-hoc lists of “must-haves.” Candidates don’t carpet-bomb every opening. Instead both sides achieve optimal outcome when they state their true preferences; the system finds the few matches that could work; humans spend their energy on chemistry and persuasion.

Recruiters may argue there is no secret sauce in this process, its obvious. But that's the point, the technology has simply unlocked an obvious solution.
In fact, such 'stable matching' is a well understood process, first studied in marriages and formalised by Gale & Shapley in the 1960's. It has been used before in recruitment. See Jeremy McEntire's Linkedin article on the subject (link below)
So, the matching process conducted by Marco requires professionalism. But there are many algorithms which can achieve this and some can do so whilst guaranteeing privacy. There will be a marketplace in Marco's, the optimal matching solution will rise to the top for each industry.
Why Player's Interests Are Aligned
Each company adopts one 'Busby', having any more duplicates effort for HR teams trying to explain their needs. Companies such as eightfold.ai already offer such agents.
Similarly, each candidate would likely have only one 'Candy'. Its a great deal of effort to converse with an agent such that trust develops to represent them faithfully. There could be many 'Candy' agents to choose from, perhaps specialised by location and/or profession.
Crucially for candidates, it is conversation which builds context and depth of understanding. Agents do conversation in a way that LinkedIn does not. Which is curious, because this blog post was inspired by recent insights on agentic AI from Reid Hoffman, founder of LinkedIn (see link at end)
Yet, Marco is the key. This is the marketplace agent and it has suddenly been enabled by these shared standards for agents to co-operate. Marco can now easily converse with or share data with any version of Candy (candidates representative) and Busby (business representative).
Since Marco is used by both candidates and businesses, it could be paid by both and trade on its reputation for fair service and preservation of privacy, which could be evidenced.
The world can accommodate many Marco's, Candy's and Busby's, there is room for specialist recruiters, no need for a central operator, the agents can talk to many other agents.
This revolution needs no-one's permission and adoption is driven by everyone's convenience. Its easier for all parties to work though intermediaries than engage in the current process. The new standards mean they can co-operate in machine time, not human time.
Co-operation becomes rational because...
Core interest | Alignment lever that satisfies it | |
Candidate / Candy | Be given a fair shot and a quick answer. | Capped preference list removes spamming; skill proofs curb exaggeration. |
Business / Busby | Hire well and fast, protect sensitive data, no swamp of CV's to trawl. | Stable-matching yields small, high-signal shortlists; progressive disclosure hides salary bands until late stage. |
Marketplace / Marco | Market liquidity, reputation, compliance. | Bias audit & immutable logs; bad actors lose reputation quota. |
People Still Hire People — The Peace Dividend
People choose a career in recruitment and HR because they want to work with other people, not become data scientists. The intent is that automation removes drudgery, but human judgement remains the final decision maker.
Recruiters and HR partners stop résumé triage and refocus on coaching, story-telling, and persuasion. Their edge is understanding nuance and selling the mission
Hiring managers & teammates can spend time on chemistry checks—pair-programming sprints, coffee chats, shadow days—because only people working together know culture fit
Candidates can finally talk about hopes and growth instead of guessing keywords. Honest motivation replaces gaming of the recruitment process
Software makes the match; people make the commitment.
Early Services Lighting the Way
Pieces of this vision already exist—they’re just isolated. The next table shows how they map
Existing service | What it already does | Which agent role it mirrors |
Voice agents: Jack interviews candidates; Jill interviews hiring teams, then makes introductions (LinkedIn) | Jack = Candy (Candidates') Jill = Marco (Markets) | |
AI sourcing finds “hidden-gem” candidates missed by keyword search, cutting time-to-interview (TechCrunch) | Marco -style matching | |
HireVue game-based assessments | Blend mini-games with AI-scored interviews; produce portable skill proofs (hirevue.com) | Skill-test micro-agents |
Eightfold AI | Generates job descriptions, maps skills, an agent to manage internal processes—essentially a Busby prototype (Eightfold) | Busby |
With MCP and A2A, these islands can interoperate, turning point tools into one seamless highway.
Conclusion
Cars only fulfilled their promise once we built roads designed to accommodate them. Open-agent standards give us the opportunity to do the same for hiring: machines coordinate, humans connect. Time to leave the muddy lane behind and enjoy the ride.
References
“50 HR & Recruiting Stats That Make You Think,” Glassdoor (2015) — 250 résumés per job (Glassdoor)
“99+ Must-Know Résumé Statistics,” Novorésumé (2025) — 80 % filtered by ATS (Novorésumé)
Anthropic, “Introducing the Model Context Protocol,” (Nov 2024) (Home)
Google Developers Blog, “Announcing the Agent-to-Agent Protocol (A2A),” (Apr 2025) (Home- Google Developers Blog)
D. Gale & L. S. Shapley, “College Admissions and the Stability of Marriage,” American Mathematical Monthly 69 (1) (1962) (Harvard EECS)
Ron Miller, “Moonhub wants to transform the way companies find job candidates using AI,” TechCrunch (Feb 2023) (TechCrunch)
Matthew Wilson, LinkedIn post introducing JackandJill.ai (Apr 2025) (LinkedIn)
Eightfold AI product page, accessed 6 May 2025 (Eightfold)
HireVue, “Game-Based Assessments,” accessed 6 May 2025 (hirevue.com)
"Job Market Inefficiencies and the Automation Arms Race", Jeremy McEntire, (Linkedin)
"The AI Use Case No One Is Talking About", 3 May 2025, Reid Hoffman -founder of Linkedin- in Interview (YouTube)
コメント