You just finished something hard. Not admin. Not email. The actual thing — the thing you were thinking about for three weeks, the project you weren't sure you could finish, the proposal that was harder than it looked. You want to tell someone. Someone who knows. Not who knows that you finished — who knows what you carried to finish.
There is no one like that nearby. You open ChatGPT. You paste the story in. It says "Great job! That sounds like a real accomplishment." You feel emptier than when you started.
The praise was not the thing you needed. You needed presence. AI cannot be present. But it can do something else — and this is where a daily routine with an AI assistant (workflow with AI) earns its place. It's an inner routine, not a task manager: it holds the attention direction of the work you've been carrying, not your calendar.
I wondered what would happen if a bot was instructed, very precisely, to not praise — and to point at the human who actually needs to be told.
Stage 6 — The energetic signal beneath the moment
Maybe the unwitnessed feeling isn't loneliness in general. Maybe it's a phase signal. You're often somewhere in the Purge phase — the stretch where the work that needs witnessing is exactly the work the old, build-alone version of you could not have made. The felt-significance of it is being cleared of I have to do this by myself, and the body hasn't found anyone to hand the new version to.
Underneath that runs the set-point. The body that broadcasts I-need-to-be-seen into the whole field at once — the Absorbing + Performance broadcast, taking in everyone's need for witness while quietly pushing its own work out into the same crowd. The applause comes back, and it lands on nothing. Perhaps the praise from a stranger doesn't sink in because praise is about the output; what you needed was witness of the process.
The mirror reading is plain here. Applause without being seen is the field returning the signal you sent: I broadcast, but I never routed. Naming who has actually watched what — one specific person — changes the broadcast into an address.
That naming is also where the Hyperindependent protective pattern gets a quiet bypass. The bot doesn't ask you to drop I-handle-it-alone by willpower. It routes the disclosure to one named person, in real words, so you never have to ask the field at large.
The optional structural overlay — the chart-level confirmation underneath — would name this an Open G center (the identity-and-direction center, the diamond at the chest on the BodyGraph (Human Design body chart) with no color — your sense of self comes from the field around you). An Open G needs to be seen in its work to know whether the work is itself; without an ongoing witness who watched the months of preparation, it can't tell whether what just shipped was the real version or a borrowed one. The same chart layer explains why Projectors (the Type built to guide and see systems, ~21% of people) live this loop most: recognition is their basic operating requirement, not a preference. A defined G center can move forward without an external witness. The energetic signal leads; the chart confirms.
Stage 7 — The question the AI asks first
The Witness Bot does not ask "how do you feel?" That's a coach's question, not a witness's. A witness does not ask about feelings. A witness sees the work and reflects what is being seen.
Instead, the opener:
"Tell me what you just finished — not what it produced, but what it actually was. What were you doing, and what made it hard?"
When the founder answers, the bot asks again — about process, not output. "Where did you almost give up?" "Which decision came slowest?" "What part of this would you not have told anyone if I hadn't asked?"
These are the questions a peer who walked alongside you would ask. The bot is borrowing the shape, not pretending to be the peer. The whole workflow with AI is built on the distinction.
Stage 8 — The workflow with AI itself
This is an inner routine — it holds the attention direction of the work, not the external structure of the day. A long-context AI surface (Claude Project, Custom GPT — anything that holds memory across weeks). At the start of each session, the bot reads back the last two or three things the founder has been working on, in their own words. The founder doesn't have to retell.
Trigger: end of work that mattered. Not admin. Not email. The thing the founder was actually thinking about.
Format: structured conversation, 5–8 minutes, one question at a time, no rush to solutions.
What the bot does: hold the context, ask process-not-output questions, refuse to praise, refuse to advise, close the session by pointing at a real human.
What the founder still owns: the work itself, the choice of who to tell, the actual conversation when it happens.
The closing move is the heart of the workflow:
"Is there anyone in your life right now who could witness what you just described? Not to give feedback — just to see. If there is, what's the first line you want to say to them?"
The bot helps the founder draft the first line. Then closes itself. The real conversation happens off-bot. (See the ask-builder — the same DNA, applied earlier in the cycle.)
A concrete session, in four lines:
9:14pm. Bot opens. Reads back: "You've been working on the consulting proposal for L for three weeks; the technical sections were the slow part." Founder: "Yes. I shipped it tonight. It's good." Bot: "What part of this would you not have told anyone if I hadn't asked?" Founder: "That I almost rewrote section three from scratch yesterday and didn't, because I trusted the version I had." Bot: "Who in your life would understand why that decision mattered? What's the first line you'd send?" Founder writes one line. Sends to a peer. Bot closes.
Stage 9 — The 7-day experiment that grows the capacity underneath
The workflow surfaces who could witness. The experiment is the slow-build that grows the actual roster.
Collecting Dots (7-day) — for seven days, intentional 1:1 coffees (or video calls) with one peer per day. The hypothesis: witness deficit doesn't get fixed by a single conversation. It gets fixed across months of dot inventory — small, repeated contact with the same small group of people who are watching your work over time.
Daily signal: one peer reached, one note about what was true between you. Week-end review: which two are candidates for Level 3 (longitudinal witness)?
(Companion: conversation layers (7-day) — once contact exists, the practice that helps the conversation actually become witness instead of polite exchange.)
Stage 0 — Return / Become
After three months, the bot becomes a reference, not a destination. The founder has 2–3 humans they can send the first-line message to. The shipping of meaningful work no longer ends in a paste-into-ChatGPT moment. It ends in a sent message, sometimes a phone call, occasionally a coffee.
It's not that the bot got smarter. It's that the bot did its actual job — pointed at the human, then got out of the way.
There's an identity layer here too. The kind of person who runs this loop is usually a Connector — the one who witnesses everyone else, abundantly, and almost never gets it back. The medicine isn't more giving. It's the asymmetry reversed: being seen, for once, in return. The bot is the receiving-practice for that.
AI cannot be a witness. It can only help you notice who can, and help you write the opening line you'll send. That distinction is the whole instrument. Hold it precisely, and the workflow with AI gives you back the part of completion the praise-loop has been quietly taking.
The work is yours. The witness is theirs. The bot is a threshold, not a room.
This pattern fits the Connector identity-state — the person who gives witness freely and rarely receives it — and reads most clearly through the Purge phase, when the work that needs seeing is the work the old self couldn't make. The optional structural overlay confirms it on the chart as an Open G center (no-color identity-and-direction center — sense of self comes from the field) and the Projector Type (~21% of people). The bridge for asking the witness for help is the ask-builder; the bridge for staying with the loop in the first place is the reflection bot. See your chart →