AI Employee 'TechHive Agent' Operates for 3 Months in Group Company's Recruitment — Human Response Time of 20 Minutes per Application Reduced to Zero, Staffing Cut from 3 to 1

Asikaze Co., Ltd. (Headquarters: Shibuya-ku, Tokyo; CEO: Daisuke Hirai) has introduced 'TechHive Agent,' a platform that assigns AI agents as 'AI employees' to perform business tasks, into the engineering recruitment operations of its group company, Ansel Technologies. The company is now releasing operational results from a three-month pilot period (April 14 to July 13, 2026).

Prior to implementation, human handling of pre-interview tasks averaged 20 minutes per application, which has now been reduced to nearly zero. The recruitment workload, previously shared among three staff members, is now managed by one human and AI agents. Initial responses to candidates are completed within a median of approximately 7 minutes—regardless of night or holiday—and immediate responsiveness has been achieved without increasing headcount. Concurrently, Asikaze has begun recruiting pilot companies ahead of the official launch.

Background: 'Every Application Interrupts Our Workflow'

Ansel Technologies operates in the IT staffing (SES) sector, where engineering recruitment is a lifeline directly tied to business growth. However, most recruitment efforts were consumed not by interviews themselves, but by preliminary desk work.

- Checking notifications whenever an application arrives via job boards - Manually transferring applicant data and drafting initial replies - Individually verifying whether applicants meet requirements, updating management sheets, and coordinating interview schedules between candidates and interviewers through back-and-forth communication

While each task takes only a few to several minutes, the biggest issue was that these were unpredictable 'interruption tasks.' Staff had to repeatedly pause their core work, and delays risked reducing candidate engagement. Despite recruitment being a 'speed determines success rate' function, the team could not balance responsiveness with focus.

Implementation: The AI Employee 'Recruitment Assistant' Takes Over Interruption Tasks

TechHive Agent (THA), developed by Asikaze, is not a single AI tool but a platform based on the concept of assigning specialized AI employees per business function. For Ansel, an AI agent specialized in recruitment—'Recruitment Assistant'—was deployed.

### [Tasks Automated by the AI Agent]

1. Initial Candidate Response — Upon detecting an application, the AI organizes applicant information, checks for duplicates, and sends an initial reply 2. Requirement Screening — Automatically cross-checks applicants against pre-defined human-set requirements (must-have criteria) and compiles results. The AI does not make evaluative pass/fail decisions; qualified candidates are handed over to human recruiters 3. Interview Scheduling — Finalizes interview times through communication with candidates and shares details with relevant parties 4. Patrol Monitoring — Every 30 minutes during business hours, the AI checks the recruitment management interface for replies or status changes and triggers follow-up actions

### [Approval Design for 'Not Acting Autonomously' (HITL)]

The core design principle of THA is human control (Human-in-the-Loop), not AI autonomy. Business actions are categorized by risk level, and the AI executes tasks only within predefined procedures and message templates approved by humans. When a decision exceeds these boundaries, the AI sends a prior approval request to the human operator, who can review the content on Slack and one-click approve, request edits, or reject. During the pilot, the AI sent 188 confirmations/reports to humans, including 50 approval requests requiring decisions—all processed with human oversight. The system is designed from the ground up to prevent the 'black box' scenario of 'we don’t know what the AI did.'

Results: Three-Month Operational Performance

- Operation Period: April 14 to July 13, 2026 (approx. 3 months) - Total Recruitment Tasks Handled: 2,107 - Initial Candidate Responses: 657 - Duplicate Application Checks: 651 - Requirement Screening: 523 - Interview Scheduling: 174 - Document Request - Status Updates, etc.: 102 - Human Work Time per Application (pre-interview): Average 20 minutes → 0 minutes - Response Speed for Initial Candidate Contact: Median ~7 minutes (90% within ~30 minutes, including nights/holidays) - AI-Substituted Work Hours (human-equivalent): ~83 hours/month (~90% of recruitment tasks excluding interviews) - AI-to-Human Confirmations/Reports: 188 (including 50 approval requests)

In the most recent fully active month (June 2026), the system processed 262 initial responses, 220 requirement checks, and 70 interview scheduling tasks in one month.

Before implementation, Ansel’s pre-interview recruitment tasks were shared among three staff members. Each new application required an average of 20 minutes—from notification checks, data entry, sheet updates, requirement verification, to scheduling emails. With current application volume (~250/month), this alone amounted to over 80 hours monthly. Including interviews (~60 minutes per session), the workload exceeded one full-time employee’s capacity, necessitating team sharing.

After implementation, all pre-interview processes are handled by the AI agent, reducing human effort per application to nearly zero. Human recruiters now only act when interview schedules are confirmed (e.g., confirmation messages, reminder setup—~10 minutes per case) and during interviews themselves. The recruitment line is now operated by a single staff member. The AI substitutes approximately 83 human-equivalent hours per month—about 90% of non-interview recruitment time.

Beyond the numbers, the biggest impact was the elimination of interruptions. Not only was human labor reduced to zero, but initial candidate responses are completed within a median of ~7 minutes, day or night. The dual goals of 'not interrupting staff' and 'not making candidates wait' were achieved. Recruiters can now focus on interviews and candidate assessment—tasks only humans can perform.

### [Comment from Recruiter (Shota Uchiyama, Recruiter, Ansel Technologies)]

'To be honest, I was skeptical at first—'Can we really trust AI with applicant responses?' Before, every day started with clearing overnight applications, and I was constantly interrupted by notifications during the day. Now, I only act when an interview is scheduled—that’s basically it. Seeing that late-night applications were handled within minutes the next morning, I’ll admit I felt a bit jealous (laughs). When decisions are unclear, Slack alerts me for confirmation, so I don’t feel uneasy about fully delegating. The freed-up time now goes into interview prep and truly engaging with each candidate. This feels like the job I always wanted to do.'

Next Steps: Pilot Program for Test Companies Ahead of Official Launch

These results were achieved within Asikaze’s own group, and THA is currently in the pre-launch phase. Asikaze is now recruiting pilot companies—particularly those in SES and staffing industries facing recruitment workload challenges. - Target: Companies struggling with recruitment workload, especially in SES and staffing sectors - Process: After business consultation, AI agents will be customized and deployed based on your current job boards and management tools - Details: Inquiries: https: