Before your best people leave, their knowledge stays.
Every company has someone who just knows — why that machine jams on cold mornings, which customer pays late but always pays, what the old system really does when you press that button. Halef uses Claude to interview them, learn how they work, and build a successor agent they review and approve. Then the whole team can ask it anything.
halef (Turkish, from Arabic) — the one who comes after; a successor who carries on the work.
The most valuable knowledge in a company was never written down.
It lives in the heads of people who have done the job for decades. When they retire or move on, it disappears — and the company pays for years to rediscover it. Wikis don't help: nobody writes them, and nobody reads them.
A whole generation of skilled operators, engineers and old-system maintainers is reaching retirement at the same time.
Most expertise is tacit. People can't write it down, but they can show it and talk about it when someone asks the right questions.
Small and mid-sized companies often have one person per critical process. One resignation can stop a line.
Six weeks from expert to successor.
Halef works the way a curious apprentice would. It asks questions, watches, takes notes and asks again until it understands. The expert stays in charge from start to finish.
Adaptive interviews
Short voice or text sessions that fit around real work. Claude asks follow-up questions about exceptions, edge cases and "why".
Learn from the work itself
The expert can add screen recordings, photos, old emails, spreadsheets and legacy code. Halef links what they say to what they do.
A playbook they approve
Knowledge becomes structured playbooks: decisions, thresholds, warning signs. Each one has a source, and the expert signs it off.
An agent the team can ask
New people ask the successor agent in Slack, Teams or WhatsApp. It answers from approved playbooks only, cites them, and says when it doesn't know.
Built with the expert, not around them.
Halef only works if veterans trust it. They are the author, not the subject. It preserves their judgment and credits them. It does not replace them.
- Explicit consent for every session and every source. Stop or delete at any time.
- Expert sign-off. Nothing reaches the team until its author approves it.
- Grounded answers only. Every answer cites its playbook. If no playbook covers the question, Halef says so instead of guessing.
- Attribution. Answers credit the person whose knowledge they come from.
- Your data stays yours. You can deploy it in your own cloud. Interview data is never used to train models.
Why now, why us.
Until recently, a machine could not run a two-hour interview, notice what was left unsaid, and come back the next week with better questions. Claude can. Long context, careful reasoning and honesty about uncertainty are what this problem needs.
"I've maintained systems where the only documentation was one person's memory. I've watched factories lose a week when that person went on holiday. I'm building Halef so that never has to happen again."
Semih, founder. MLOps engineer. Has built and maintained operations software for manufacturing, construction and retail SMEs.
Who at your company can't be replaced?
We're onboarding a small group of design partners in manufacturing, construction and distribution. Tell us about the role and the knowledge you want to keep.
info@ssilistre.tech