The AI layer that makes you better at your work, not dependent on it.
TAOS sits between you and any LLM. It accelerates the work you are already strong at, coaches the work you are growing into, and keeps the skills you care about from quietly atrophying under AI use.
Free trial, cancel in one click. Works in Claude, Claude Code, ChatGPT, Gemini, Cursor & Codex. Your profile is portable markdown: no lock-in.
From our 2026 pilot.
Three steps. Then it just runs in the background.
No new app to live in. TAOS rides along inside the AI you already use.
Connect it to your AI
One command in Claude Code, or a connector URL in Claude, ChatGPT, Cursor and the rest. One profile, every tool.
Tell it about you
A short conversational assessment maps where you are expert, where you are growing, and which skills you want to protect.
Get a coach, not an autocomplete
From then on, every task is matched to the right mode: accelerate, coach, protect, or hand the call back to you.
From a five-minute chat to a sharper you
Tell it about your work, get a profile, then coaching that builds the skill instead of doing it for you.
Your skill and your output quality climb together.
Five modes, matched to you
Your profile tells the AI which mode to use for which kind of work, so it gets out of your way when you are expert and slows you down on purpose when the thinking is yours to do.
- Automate: AI executes and annotates. Low friction, mechanical work.
- Augment: AI accelerates and challenges you in your expert domains.
- Coach: AI scaffolds and questions. For skills you are actively developing.
- Protect: AI forces cognition to keep at-risk skills sharp.
- Hands-off: AI surfaces the decision and lets you make the call.
See exactly where AI should push you, and where it shouldn't
The quick assessment takes about five minutes (deepen it later with the full conversational one). At the end you get a personalised map of your strengths, growth areas, and at-risk skills.
Your full profile, scores & coaching
Start a 3-day free trial. Cancel in one click before it renews.
Build my profile →Unguarded AI makes you faster today and worse over time
Three independent studies, one finding: AI without guardrails causes deskilling. TAOS is the guardrail layer.
Accuracy drop on no-AI exams
RCT of ~1,000 students. Unguarded AI users scored 17% below the no-AI control on a closed-book exam. Guardrails fully neutralised it.
Bastani et al., PNAS, 2025.
Accuracy outside the frontier
RCT of 758 BCG consultants. AI was 25% faster, but 19 points less accurate on tasks outside the model's competence.
Dell'Acqua, Mollick et al., HBS & BCG, 2023.
Cognitive debt is measurable
EEG essay study. Users who later switched off AI showed weaker neural engagement than the no-AI control. Over-reliance leaves a trace.
Kosmyna et al., MIT, 2025.
A personal upgrade. An organisational safeguard.
Grow into your role while you ship more.
Two AI risks in your workforce. TAOS was created to solve both.
Install in your tool
Same hosted endpoint, Google sign-in, one portable profile. Pick the client you already use.
One command in your terminal. OAuth opens in your browser on first use.
claude mcp add taos --transport http https://taoshq.com/mcp
For ambient coaching from turn one (a SessionStart hook) plus local-filesystem skills, install the TAOS for Claude Code plugin from the GitHub marketplace.
Then ask Claude: Run my TAOS assessment
Add the TAOS plugin to Cowork from its public marketplace, then start coaching. No org admin needed.
- In Claude, open Customize, choose Add marketplace, then Add from a repository, and paste this owner/repo:
angelo-leone/taos-connector
- Click Sync to load the marketplace.
- Install the TAOS plugin from it. Sign in with Google when your client asks, so the connector can reach your profile.
- Say
Run my TAOS assessment, then bring a real task and let it coach you.
Your profile and the coaching method stay on the TAOS server; the plugin only connects you to them. Rolling out to a team? A Team or Enterprise owner can deploy it for everyone: talk to us.
- Claude desktop or claude.ai → Customize → Connectors → + → Add custom connector.
- Name it TAOS, paste the endpoint, click Add. Google sign-in runs automatically.
- Ask Claude
Run my TAOS assessment.
https://taoshq.com/mcp
Prefer a local install for Claude Desktop? Download the desktop extension (.mcpb) and double-click it. Your profile stays on your machine.
- Turn on Settings → Apps → Advanced → Developer mode (and leave it on; custom MCP apps need it).
- Settings → Apps → Add new app, name it TAOS, paste the endpoint, set Authentication to OAuth, and finish Google sign-in.
- Ask ChatGPT
Run my TAOS assessment.
https://taoshq.com/mcp
Gemini does not support custom MCP connectors yet, so TAOS works as a paste-in Gem. Build your profile on the web first:
https://taoshq.com/assess
- On your dashboard, click Download → Gemini.
- Paste the bundle into a new Gemini Gem called TAOS Coach.
Any MCP-aware client (Cursor, Windsurf, Codex, Gemini CLI) can connect over HTTP with OAuth:
https://taoshq.com/mcp
See the install guide for per-client specifics.
Prefer to self-host? Talk to us →
What early users said
Real, unedited feedback from people using TAOS. Brand name shown in brackets.
Common questions
Does using AI actually make you worse at your job?
Unguarded use can. In a randomised trial of around 1,000 students, people using an unguarded AI tutor scored 17% below the no-AI control on a closed-book exam, while a guardrailed version of the same tutor erased the gap (Bastani et al., PNAS, 2025). In a trial of 758 BCG consultants, AI users were 25% faster but 19 points less accurate on tasks outside the model's competence (Dell'Acqua, Mollick et al., 2023). The damage is not automatic. It depends on whether the tool lets you skip the thinking.
What is AI skill atrophy?
The gradual loss of a skill you used to practise, because AI now does it for you. It is invisible while you have the tool, because output quality holds up. It shows when the tool is taken away, or when the task falls outside what the model is good at and you can no longer tell. TAOS exists to keep that from happening to skills you said you cared about.
How is this different from telling ChatGPT to ask me questions first?
A prompt applies the same friction to everything, which is why people stop using it by Wednesday. TAOS rates your expertise per domain and varies its behaviour: it accelerates work you are already good at, and only slows down where slowing down buys you a skill. That judgement persists across sessions and across Claude, ChatGPT, Gemini, Cursor and Codex, so you are not re-explaining yourself every morning.
Does it work with the AI I already pay for?
Yes. Claude, Claude Code, ChatGPT, Gemini, Cursor, Codex, and any MCP-aware client. One profile follows you across all of them, so you are not locked into a single vendor.
Is it free?
Connecting TAOS to the AI you already use takes a couple of minutes. The full profile, dashboard and coaching come with a 3-day free trial (card at checkout, one-click cancel before it renews). After that it is a flat price.
Will the coaching slow me down?
It does not coach everything. More than half of real tasks land in your expert domains, where TAOS goes fast and challenges the output rather than asking you questions. Coaching is reserved for the places it pays back. Some pilot participants finished their work faster than before. And when you just need speed, switch to ship mode for that task.
What happens to my data?
Your profile and lightweight usage signals are stored so coaching can adapt. You can export or permanently delete everything at any time. GDPR-ready, with a self-host option. See the privacy policy.
Can my company deploy it?
Yes. Enterprise gets aggregate-only dashboards (no individual surveillance), audit logging, and member management; SSO, self-hosting, and EU residency are available on request. Talk to us.
Get better at your job. Starting now.
Install the connector in a couple of minutes and feel the difference on your next real task.