toishi 砥石whetstone
Sign inSign up

See your team's skills fade before your product shows it.

With AI, a team can ship faster than ever while it slowly stops understanding its own code. Most leads find out late, from a bad production bug or an outage. Toishi shows the early signs, in team totals, and gives everyone time to practise.

Team pricing

The team's skills, never one person's.

Leads see each skill the team works in: how much of it the team still types by hand, and whether that is going up or down over the last four weeks. For each repository, they see how much of it the team still writes by hand.

When a skill is fading, give the whole team an exercise for it.

Skills

Per skill: how much the team types by hand, and where it is moving. Last four weeks.

SkillTyped by handvs. 4 weeks beforeAll linesDev-days
typescript
32%
4 pts31,200164
rust
25%
6 pts12,48071
sql
14%
7 pts4,35038
css
58%
1 pts3,91052
python
16%
4 pts2,64019

What leads see

Overview
How much the team still writes by hand, and how sharp it is.
Skills
For each skill, how much the team types by hand and where it is moving.
Trends
The team over time, by day or by week.
Repositories
How much of each repository the team still writes by hand, and the skills it calls for.
Recommendations
What the team could practise, worked out from its totals by fixed rules.
Assignments
Exercises for the whole team. Members see them in the app, and nobody's progress is tracked.

Practice needs your backing.

Under pressure to ship with AI, few developers take time to practise on their own. When the team plan pays for it, an hour or two a week is part of the job.

Seniors keep the skills you pay them for, and the ones who shape your product stay able to judge what the agents write.

Juniors who learn with AI from day one.

We once interviewed a developer who could not explain how the product they had built worked, or which technologies it used.

Toishi shows which skills the team practises and which it hands off, so onboarding can aim at the gaps.

What the research says

Early studies, with small groups. They point the same way: working code from AI is not the same as a team that understands it.

Built so it cannot rank people.

Each developer's stats stay encrypted on their own machine. Once a day, the app sends a fixed report, split into two halves for two servers run by different parties. Neither server can read a report on its own. Together they only give out totals.

Groups of fewer than five people show nothing, and there is no page about a single person.

How privacy works

Try it on your own work first.

Create my account