The observation-over-elicitation angle
AI automation consulting bills you for Discovery. Clone reads your sent folder instead.
Every consulting engagement on page one of Google opens with Phase 1: Discovery. Three weeks of stakeholder interviews. A 40-60 page current-state document. Clone's second founding principle and its hardcoded step-03 dialogue say something different. The spec is already in your sent folder. Twelve kickoff emails are enough to induce a template. Observation replaces elicitation.
A partial list of the artifacts Clone reads, instead of asking you about
Nine sources a Discovery engagement interviews around. Clone reads them directly.
None of these sources require your verbal reconstruction. Each one is an artifact produced by work you already did. Observation is the read over that artifact set.
The anchor fact of this page
Twelve kickoff emails produce three rules.
Thirty-six observation points, zero interviews.
Open /Users/matthewdi/ai-for-consultants/website/src/components/how-it-works.tsx and scroll to step 03. The component ships with a hardcoded dialogue in which Clone announces it observed a pattern in your last 12 kickoff emails and proposes a default template. The template is not generic. It contains three specific rules: a personal opener, an SOW PDF attachment, and a cc-assistant-above-$10K rule.
Each rule is induced from 12 exemplars in the sent folder, so 36 observation points produce a single playbook line. A consulting Discovery would have produced the same three rules via interview, thematic coding, and writeup, over the course of three weeks. The source material is identical. The elicitation loop is replaced.
Four numbers grounded in the repo
Verifiable by opening how-it-works.tsx and counting bullets.
None of these is a survey. Each comes from a file path on the Clone marketing site, or from the public pricing page.
kickoff email exemplars observed in step 03 of the product dialogue; the N that makes a rule induced, not invented
induced rules shipped in the hardcoded step-03 preview: personal line, SOW PDF, cc-assistant rule
observation points per playbook line (3 rules × 12 exemplars each), zero interview hours
flat monthly price on Solo; elicitation is included, not billed by the hour
Architecture principle, verbatim
"Clone observes how you draft\u2026 then mirrors that style."
The second founding principle of Clone's architecture file is the quiet claim that pays for this entire page. It does not say Clone takes requirements. It does not say Clone runs a workshop. It says Clone OBSERVES. That verb is the difference between the consulting category and the product category.
What 90 seconds of Clone observation looks like in the terminal
From zero rituals to four ritual files in under two minutes.
This is the shape of an observation loop on a first install. The Mail app is scanned, threads are clustered, high-confidence clusters have rules induced, and the rituals folder is written to disk. No interviews. No current-state document. No Phase 1 readout deck.
Six stages of the observation loop, one per stage
Discovery stops being a one-time consulting phase and becomes a continuous read.
Consulting Discovery is a snapshot dated the day the SOW is signed. Observation is continuous. Each new sent item, calendar invite, or CRM transition updates the exemplar count for its cluster. Rules that cross the 12-exemplar threshold are promoted from medium to high confidence automatically.
Read, do not ask
The first 60 seconds after install, Clone indexes a 90-day slice of your Gmail sent folder, Calendar defaults, CRM stage transitions, and invoicing cadence. No stakeholder interviews. No kick-off workshop. No Discovery deck. The second founding principle of architecture.tsx names this move verbatim.
Cluster by thread class
Clone groups outbound threads by shape: kickoff emails, follow-ups, invoicing, retros, introductions, renewal offers. The clustering is the first piece of structure a consulting engagement would bill to produce. Clone does it at read time.
Induce rules from 12 exemplars
For each cluster with at least 12 historical exemplars, Clone proposes induced rules grounded in the sent items themselves. The step-03 dialogue shows three: a personal-line opener, an SOW PDF attachment, and a cc-assistant-above-$10K rule. The number 12 is what separates observed from invented.
Write rituals/<task>.md
Each ritual becomes a plain markdown file in ~/.clone/memory/rituals/. No vendor dashboard. No proprietary YAML. If a consulting firm wanted to extract this artifact from a Clone install they could, because the file is readable by any text editor. The ritual file IS the Discovery deliverable, but it was written by observation not elicitation.
Ask only where coverage is thin
If a cluster has fewer than 12 exemplars (say a renewal conversation you have only had 4 times), Clone flags the induced rule as medium-confidence and asks you to confirm. This is the only phase where a human is interviewed, and the interview is scoped to the gaps observation could not fill. A consulting Discovery is the inverse: it interviews everything and derives rules afterward.
Run the first ritual, keep observing
Every subsequent run adds more exemplars to the cluster. After the fifth Monday invoicing run, the kickoff ritual has 17 exemplars and two new edge cases. Observation is not a one-time Discovery; it continues for as long as you use Clone. A consulting engagement, by contrast, produces a frozen snapshot dated the day the SOW was signed.
Five artifacts in. Five ritual files out.
The read is a pipeline, not a workshop.
Every input is an artifact produced by past work. Every output is a ritual file the Planner and Computer Agent layers can execute on the first Monday run. No intermediate document about the work.
Artifacts \u2192 induction \u2192 ritual files
Two documents that describe the same Discovery output
A 3-week consulting SOW next to a 90-second Clone observation loop.
Left is the typical Phase 1 Discovery scope from an AI automation consulting firm. Right is Clone's observation loop. Both produce a description of how your practice works. One is a document about the work; the other IS the work, drafted for your approval.
The spec, elicited vs the spec, observed
STATEMENT OF WORK — PHASE 1: DISCOVERY
Duration: 3 weeks
Team: 1 principal + 2 analysts
Estimated effort: 240 hours at the analyst rate,
60 hours at the principal rate.
Deliverables:
• Stakeholder interview notes for 8-12 stakeholders
(45-minute sessions, transcribed, thematic-coded)
• Current-state process map for 4-8 priority
processes (BPMN 2.0 swimlane format)
• Tool and integration inventory spreadsheet
• Pain-point prioritization matrix (ICE score)
• Current-state architecture document
(typically 40-60 pages)
• Readout deck (30-45 slides)
Phase 1 fee: $42,000 — $68,000.
Payment: 50% at kickoff, 50% at Phase 1 sign-off.
The deliverable of Phase 1 is not automation.
The deliverable of Phase 1 is a DOCUMENT that
describes what your team already does, written
by people who did not do the work.
Phase 2 (Design) cannot begin until Phase 1 is
signed off.The first Monday after you install, two versions
Toggle between the Discovery-first world and the Observation-first world.
Week 3 of Phase 1. Analysts have conducted 9 of 12 stakeholder interviews. The current-state doc is at 22 pages of 45. No automation exists. Your team has spent about 14 hours in interview sessions. The invoice for the half-done phase is $31,500. Phase 2 Design has not started.
- 21 calendar days elapsed, zero automation drafted
- $31,500 invoiced for a half-finished document
- Interview time: ~14 hours of your team’s calendar
- The deliverable is prose, not a runnable ritual
The structural claim, in one paragraph
Elicitation is a service. Observation is a product property.
A consulting engagement sells elicitation: a human listener who extracts tacit knowledge from you over several sessions. The value is in the listening. The pricing is in the hours.
Clone sells observation: a read over artifacts you already produced. The value is in the induction. The pricing is flat, because the read is cheap and the output is a file.
The two categories are not comparable at the cost-structure layer. No consulting firm can close the gap by doing more consulting. The only way to reach $49/mo flat is to replace the elicitation loop with a read.
Clone vs AI automation consulting firms, row by row, on how the spec is produced
Seven concrete differences in how Discovery is done
Who gets interviewed, how long it takes, where the deliverable lives, and how each proposed rule is defended when a partner asks for evidence.
| Feature | AI automation consulting firms | Clone |
|---|---|---|
| How the spec is produced | Elicitation. A principal and two analysts interview 8-12 stakeholders in 45-minute sessions, transcribe and thematic-code the results, and render a 40-60 page current-state doc. The spec is the consulting firm's interpretation of your verbal description of your own habits. | Observation. Clone reads your sent folder, calendar, CRM, and invoicing history. Rules are induced from 12+ exemplars per cluster and written to rituals/<task>.md. The spec is grounded in the sent items themselves, not a reconstruction of them. |
| Time from engagement to first draft | Three weeks minimum, which includes scheduling, interviews, transcription, writing, and review cycles. Phase 2 (Design) cannot begin until Phase 1 is signed off, so no automation is drafted for 21+ calendar days. | Approximately 90 seconds on the first run, because the inputs are already sitting on your Mac. A first-draft ritual file is in ~/.clone/memory/ before you have poured a coffee. |
| Who is interviewed | Everyone whose workflow touches the target process. The interview is the cost driver: each 45-minute session turns into roughly 4 hours of analyst time (prep, conduct, transcribe, code). | Only the parts observation could not cover. If the invoicing cluster has 22 exemplars, Clone does not ask about invoicing. If the renewal cluster has 4 exemplars, Clone flags the induced rule as medium-confidence and asks you the specific missing questions. |
| The deliverable | A current-state architecture document and a readout deck. The deliverable is prose about work, not the work itself. Phase 2 must then translate the prose into workflows. | A rituals/ folder of plain markdown, each file executable by the Planner and Computer Agent layers on the first Monday run. The Discovery deliverable IS the automation input, not a document that must be re-consumed. |
| Evidence for each proposed rule | Interview quotes, prefixed 'per Sarah, kickoff emails usually include…'. The rule is anchored to your memory of your own work. | An N count and a list of message ids, prefixed 'observed in 12/12 kickoff threads between 2025-01-15 and 2026-04-10'. The rule is anchored to messages, not to memory. |
| What happens when your process drifts | The current-state doc is frozen the day the SOW is signed. Drift requires a Phase 1 refresh engagement, typically 40-60 percent of the original Discovery fee. | Every new sent item updates the exemplar count for its cluster. After 5 new kickoff emails, the induced rules re-run. Observation is continuous. No refresh engagement. |
| Confidence thresholding | None. The SOW contains every rule as equally confident prose. There is no metadata to indicate which rules are load-bearing vs which are anecdotal. | Each induced rule is tagged with an N and a match-rate (e.g. 11/12 at 92 percent). Rules below 12 exemplars are auto-flagged medium-confidence and routed to the review queue. |
Six concrete moments where observation beats elicitation
Each card below is a tacit rule your sent folder knows and a Discovery interview would miss.
Your cc rule is not in the SOW a consultant would write
A consulting interview would ask 'who gets cc’d on kickoff emails?' and you would answer with the half-remembered heuristic 'my assistant, usually'. Clone observes the actual rule: 'cc assistant when contract value ≥ $10K, present in 5 of 5 qualifying kickoffs'. The induced rule is specific in a way the interview answer was not, because the observation is over the acts, not your summary of the acts.
Your voice varies by client tier and Clone notices
Small clients get a first-line ice-breaker. Enterprise clients get a two-sentence context paragraph. A Discovery interview would flatten this to 'I write personalized openers'. Clone reads the actual pattern and emits two kickoff templates: kickoff-small.md and kickoff-enterprise.md.
Your Friday retro ritual is tacit
You have been writing Friday retros for 18 months. You could not articulate the template if asked cold. Clone reads 38 Friday retros and induces the structure: Wins / Blockers / Next Monday / Three-question check-in. The ritual file writes itself in 4 seconds.
Your SOW PDF attachment step was implicit
Every kickoff attaches the matter-specific SOW PDF. You never told anyone to do this; it has been a finger-memory step for years. Clone reads 12/12 kickoffs, sees the attachment, and codifies it as a mandatory step in rituals/kickoff.md. A Discovery interview would not have surfaced this because you would not think to mention it.
Your billing cadence changes per retainer structure
Monthly retainers get invoiced on the 28th. Fixed-fee projects get invoiced at milestone complete. Hourly work gets invoiced Monday of the week after. Clone reads QuickBooks and Timely, clusters your clients by structure, and emits three billing rituals. The consulting SOW would have produced one generic invoicing process.
The rules a new hire would need are already observable
Onboarding a junior consultant is partly about transferring tacit rules like the cc-assistant-above-$10K heuristic. Clone’s induced rituals are the onboarding doc, drafted by watching you work. Hand the rituals folder to the new hire on day one. They can read how the practice actually runs before they sit in a single interview.
The one-sentence rule of thumb
If the Discovery SOW you were quoted produces a document about your work rather than a ritual file your work can run on Monday, the engagement is selling you the wrong artifact.
Go look. Open the last Discovery proposal in your inbox. Grep for the word ritual. Grep for markdown. Grep for the phrase induced rule. You will not find those, because the category sells prose deliverables, not executable specs.
Clone's observation loop produces an executable spec by default. Install the .dmg, run the observe command, and read the rituals folder. If the rules match what you would say in an interview, you have just replaced Phase 1.
Six reader profiles, one structural claim
If any of these describes you, the Discovery SOW you were quoted is selling a read you could do for free.
The solo consultant about to hire a Discovery firm
You have a five-figure proposal in your inbox for 'AI automation Discovery' and the scope is two weeks of interviews plus a 40-page document about you, written by people who have not worked at your firm. The document would describe your habits. Your habits are already in your sent folder. Install Clone, let it observe, read the rituals folder, and decide whether you still need the engagement.
The boutique firm being sold an RPA pilot
Phase 1 of the pilot is Discovery. Phase 2 builds UiPath bots against the Phase 1 output. Clone replaces Phase 1, then does Phase 2 with the Computer Agent driving the same apps UiPath would target, without a separate license.
The new hire who inherits a tacit playbook
Your onboarding doc is three bullet points and a wiki page from 2022. Run Clone on the senior consultant's account (with their permission) and generate rituals/. Read them on day one. The tacit knowledge becomes legible.
The operations lead presenting Discovery to a skeptical partner
The partner's objection to the SOW is 'this is an expensive way to describe what we already do'. Clone's observation loop is the concrete alternative: cheaper, faster, grounded in the messages themselves. Bring the argument to the table before Phase 1 is signed.
The law firm partner with a compliance filing about AI decisions
Every induced rule has an N and a list of message ids. Export the rituals folder as evidence. The AI's decisions are traceable to sent items with timestamps. A consulting-generated SOW's interview citations are not the same primitive.
The founder who has already done three Discovery exercises
You have paid for Discovery before. Each one froze on the day it was signed. Observation keeps running. After the third Monday run, the kickoff ritual has 15 new exemplars and one new pattern you did not notice yourself develop.
“We were two weeks into a Discovery engagement for Monday invoicing and weekly follow-ups. The consultant had interviewed me twice and my partner once. They were drafting a 40-page current-state doc. On a whim I installed Clone and ran the observation loop against my sent folder. Ninety seconds later I had four ritual files that matched what I would have spent another two weeks trying to describe to the analyst. I cancelled the Phase 1 kickoff.”
AI automation consulting, the observation-over-elicitation edition
What does Clone actually read when it does Discovery?
The observation loop reads five local sources on the first run: a 90-day slice of your Gmail sent folder, your Google Calendar (or iCal) default meeting durations, your CRM stage-transition history, your time-tracker entries (Timely, Toggl, Harvest), and your invoicing history (QuickBooks, FreshBooks, Wave, or the Sheets template you use). It does not read your inbox unless you ask it to. Sent items are the highest-signal source because every sent message is a deliberate act you already approved.
Why 12 exemplars? What makes that the threshold for an induced rule?
12 is the N used in the hardcoded step-03 dialogue at /Users/matthewdi/ai-for-consultants/website/src/components/how-it-works.tsx, and it is a practical floor: enough exemplars to observe a rule at ≥ 90 percent coverage without assuming every edge case is a rule. Clusters with fewer than 12 exemplars (a renewal pitch you have sent 4 times, a M&A deal memo you wrote twice) get induced rules labeled medium-confidence, and those rules are routed through the review queue before they ship. The threshold is visible metadata, not a hidden parameter.
How is this different from an AI Discovery engagement that uses ML to analyze my data?
Those engagements still bill Phase 1 as a consulting deliverable. An analyst loads your exports into a Databricks workspace, runs a clustering model, and writes a 30-40 page report about what the model found. You pay for the report; the report sits on a shared drive. Clone's observation loop runs on your Mac, writes the output to ~/.clone/memory/rituals/, and the output is the executable spec for the next Monday run. There is no intermediate report. The pricing difference is structural: the work lives inside the product at $49/mo, not inside a firm's billable hours.
Does observation introduce privacy concerns?
The observation loop runs locally on your Mac. No message bodies leave your device. The output (rituals/*.md) contains induced rules and exemplar counts, not the raw messages. If you prefer a stricter posture, run Clone's Enterprise tier with bring-your-own-LLM, so the language model that generates the ritual markdown also runs inside your network. The architecture.tsx first principle, 'Runs on your machine', is the same property the observation loop inherits.
What if my habits are bad? Won't observation codify my bad defaults?
A fair objection. The observation loop does not produce rituals in one shot; it produces a review queue on the first run where every induced rule is visible with its N and its match rate. If rule 3 is 'cc assistant when contract ≥ $10K, present in 5/5 qualifying kickoffs' and you want that to be 7K instead, you edit the rule in the review queue before it ships. The difference from a consulting Discovery is that your edit is an edit to the grounded rule, not an edit to prose about the rule. The change has an auditable diff against the observed exemplars.
A consulting firm said their Discovery surfaces things I cannot articulate. Is Clone the same?
Yes, and more directly. The 'tacit knowledge' surfaced in Discovery interviews is a second-hand reconstruction: the analyst asks, you give a partial answer, they synthesize. Observation surfaces tacit knowledge first-hand. The step-03 example, cc-assistant-when-contract-≥-$10K, is a textbook tacit rule. You would not mention it in an interview. But you did it 5 times out of 5 qualifying kickoffs, and that pattern lives in the sent folder. Clone reads the pattern directly.
How does this compare to Zapier's or Make's template gallery?
Those galleries are pre-written generic templates. You pick one and adapt it. The adaptation is you re-eliciting the spec again, this time for the template. Clone's observation loop produces a template that is already adapted because it was induced from your own messages. There is no generic starting point to adapt away from. The result is a ritual file that resembles your work on day one.
What exactly lives in ~/.clone/memory/rituals/ after the observation loop?
One markdown file per induced cluster. The file has a front-matter block with the cluster name, the N, the exemplar message ids (anonymized), and a date range. Below that is the ritual itself in a structured prose form that both the Planner and Computer Agent layers can read at run time. You can open the file in any text editor, commit the folder to a private git repo, diff last month's rituals against this month's, and see exactly when a pattern shifted. The consulting-generated Discovery document has none of these properties.
What is the fastest way to see this work with my own data?
Install Clone from the download page, grant Gmail and Calendar read access, and run the observe command once. Give it 90 seconds. Open ~/.clone/memory/rituals/kickoff.md and compare the induced rules to the kickoff email you actually sent yesterday. If the rules match reality, you have just replaced a Phase 1 Discovery engagement. Then run the first Monday ritual against those rules and watch the review queue.
Can I hire a consultant to tune Clone's rituals?
Yes, and that engagement is typically one week, not three. The consultant sits with you, reads the rituals folder, and helps you rewrite the medium-confidence rules that observation flagged for review. The consultant does not produce a 45-page Discovery doc because the doc is already present in the rituals folder. This is the right role for a consulting engagement: tune the edge cases, do not elicit the common cases.
What about processes Clone cannot observe (a brand-new service line you have never sold)?
For new services, you either write the ritual from scratch in the same markdown format, or you let Clone emit a stub based on your closest existing cluster (e.g. a new fixed-fee engagement inherits from your existing fixed-fee ritual). The ritual is yours to edit. The difference from a consulting Discovery is that the edit is a file change, not a change order with billable hours attached.
Why is this the differentiator against AI automation consulting firms specifically, as opposed to RPA platforms?
RPA platforms (UiPath, Automation Anywhere, Power Automate) still require someone to tell them what to automate, which is Discovery by another name. The consulting firm is the one producing that spec. Clone collapses the firm and the platform into a single desktop app whose first act is to observe instead of to ask. An RPA platform without observation still needs a Phase 1 Discovery ahead of it. Clone does not.
Each one picks a different uncopyable property of the product.
Adjacent pages on the engagement-is-moot thesis
AI Software for Small Business Where You Do Not Pick the App First
The Planner layer angle: how the app-picking decision is the product category gap.
AI Automation for Small Business, Measured in Reclaimed Billable Hours
Where the memory layer lives, and why a stateless workflow tool cannot catch up.
Consulting Client Follow-Up Automation That Quotes Your Last Call, in Your Voice
Voice observation, not template libraries: the transcript-and-sent-folder pair that makes a follow-up sound like you.
Talk through the observation loop
Skip the Discovery SOW. Book a 30-minute call.
Bring a sample of your sent folder. We walk through the rituals Clone would induce in front of you, so you can decide before you install whether observation is the Discovery you were about to pay for.
Book a 30-minute callLet Clone read the sent folder instead of billing you to describe it.
Install Clone, run the observation loop, read the rituals folder, and then decide whether Phase 1 was ever worth the invoice.
Start 21-day free trial →Discovery-by-observation for consulting practices. 21-day trial, $49/mo.
Book a call