Bench

An agency runs on decisions, and most of them never get written down. Why a candidate was turned down, what a client really wants, which fee was agreed. It lives in someone's head, a Teams thread or a personal AI account, and when they leave, it leaves with them.

Meanwhile consultants paste CVs and client notes into personal AI accounts because it helps. That is candidate data leaving the agency with no record and no way back.

One secure brain the whole agency shares. It connects to the places consultants already talk: Teams and Slack, Outlook and Gmail, call transcripts, AI chats and CRM notes. Shared skills pick out each decision, match it to the right candidate, client and role, and file the why next to the what.

Consultants work in the agency's own Claude or ChatGPT account. They ask the brain a question, or tell it what happened and let it update the CRM, the notes and the team in one go.

Every engagement starts with a look: three to four weeks with the agency, turning its people, data and time into one ordered list. Then Bench builds the most useful thing first, checks it actually helps, and only then moves on to the next.

Bench launched in October 2026 at recbench.co. The decisions stop leaving with the people who made them, and every consultant asks the same memory.

There is no lock-in. Code, skills and documentation are handed over as each phase is paid for, and the brain works with Claude, ChatGPT or whatever comes next.

Nothing changes until the consultant says yes. Every update is shown first and logged with who made it and why, and anything the brain is unsure about is checked with the consultant.

The brain decides what each person can see, hands the AI only what the question needs, and hides names when the task does not need them.