A line-drawing read of the 2026 consulting stack
Yes, AI is coming for consultants. It is coming for the invoicing tab, not the judgment call.
The fear-industry headline treats consulting as a single monolith getting replaced. The vendors actually building in the category tell a much more specific story. Read one file on cl0ne.ai, src/components/features.tsx, and count what the six feature cards automate. Every one is back office. Not one touches discovery, strategy, framework, proposal pricing, or the recommendation itself. The line between the half that gets automated and the half you still own is drawn, verifiably, in source code.
The argument in one paragraph
Carve the job at the joint.
Admin is automated. Judgment is not.
Consulting is two jobs welded together. The first is judgment: sitting with a client, hearing the problem, picking a framework, pricing the engagement, saying out loud what you would do. That is what the invoice is for. The second is operations: timesheets, invoicing, transcripts, CRM hygiene, follow-up drafting, dashboard refreshes, the weekly round of paperwork that eats 10 to 15 hours a week on top of the billable work.
The headline "AI is coming for consultants" treats those two jobs as one target. They are not. The ops half is a formal, repetitive, state-transition problem, exactly the kind an AI operator was built for. The judgment half is a conversation in a room, a decision with a human name on it, and a relationship that gets renewed on trust. One of those halves is getting automated in 2026. The other is not.
The next section grounds that claim in a specific file on the shipped cl0ne.ai marketing site. Count the feature cards. Read the titles. The line is drawn.
The uncopyable detail
Six feature cards. Zero of them touch discovery, strategy, proposals, or the recommendation itself.
Every feature in the file is back-office ops. No feature is a "strategy generator," a "framework library," a "proposal pricing advisor," or a "discovery call replacement." The file is a product-level decision about which half of a consultant's week this product touches and which half it walks around. A grep of the file makes the split unambiguous.
The full feature array below. Each comment is mine, marking which part of the consulting workflow is automated by that card and which part is not.
By the numbers
Four numbers that reframe the headline.
What the six cards actually automate
The half of your week the vendors actually target. Every card is ops.
Each card below maps directly to one of the six objects in features.tsx. The line numbers in the descriptions point to the exact code you can read for yourself. This is the ceiling of what the product automates. There is no secret second page.
Timesheets reconciled and invoices drafted
Clone reads your Timely or Harvest tab, applies the right rate per engagement, and drafts the invoice in QuickBooks or FreshBooks. You approve. You send. features.tsx lines 13 through 21.
Zoom transcripts filed against the right contact
Every call transcribed by tl;dv, Fireflies, Otter, or native Zoom lands in the right HubSpot, Pipedrive, or Folk contact with next steps tagged. Lines 32 through 37.
Follow-up emails drafted in your voice
Clone writes the follow-up from the transcript plus your prior emails. You read, tweak, press send, or schedule. Lines 38 through 45.
Kickoff packets assembled on a signed SOW
Drop the signed proposal into a folder. The shared workspace, the kickoff agenda, the calendar invite, and the welcome email all arrive before your coffee is done. Lines 22 through 28.
Dashboards assembled from Sheets, CRM, and invoicing
Utilization, outstanding invoices, upcoming renewals, pipeline. Refreshed every morning. You read them; you still make the decisions. Lines 46 through 53.
What the file deliberately walks around
The half no feature card touches. Still yours, and it is the half clients pay for.
None of the five items below are automated by any card in features.tsx. Architecture principle 2, lines 51 through 53 of architecture.tsx, is the vendor's own language: "your workflows, your voice." The framework, the conclusion, and the conversation are still the pen that signs the engagement letter.
Discovery conversations
The 45 minute call where you hear the actual shape of the problem. No feature in features.tsx tries to replace it. A transcript of it, however, is logged automatically.
Framework and scoping decisions
Which model to apply, how to stage the engagement, what to leave out of scope. The parts clients pay real money for. No feature card touches this.
Proposal pricing and tradeoff conversations
The judgment call on what a piece of advice is worth and how to defend that number in a live conversation. Still yours, and the proposal template you wrote yourself is the artifact Clone drafts from.
The recommendation itself
The sentence you say at the end: 'here is what I would do.' Clone assembles the evidence you read. It does not write the conclusion. Architecture principle 2, lines 51 to 53: 'your workflows, your voice.'
The relationship and the room
Client trust, body language, hard conversations, retention after a bad quarter. The product runs on your machine; the client still wants you in the room.
The shape of the automated half
Your admin tabs go in. Drafts you approve come out. You are still the signer.
On the left, the tabs you have had logged in for months. In the middle, a Mac running Clone, reading an English sentence you typed. On the right, the same tabs, one tick later, with a queue of drafts awaiting your review. Nothing reaches a client that you have not approved.
A single Friday ritual, end to end, across the tools you already own
Same consultant, two weeks
Friday, 4pm. Before Clone, and after.
Friday, 4:00pm. Three clients to invoice, one follow-up overdue, a dashboard to refresh.
You close the client call at 3:50pm. You open Timely, reconcile the week. You switch to QuickBooks, draft three invoices, copy the right rate for each. You send from Gmail. You switch to HubSpot, paste a note against each contact. You flip to the Google Sheet and update the pipeline row by hand. You realize at 5:45pm you have not done the follow-up for yesterday's discovery call. You write it in a rush. You close the laptop at 6:30pm, tired of admin.
- Three apps, eleven tabs, no automation
- 45 minutes of invoicing, 20 minutes of CRM hygiene
- Follow-up written in the last two tired minutes
- Weekend shadowed by the Monday inbox
- The billable judgment of the day buried under ops
Fifty years, one pattern
Every time a wave "came for consultants," it came for the admin function.
The same story, told five times. Each wave consolidated an ops role into a tool. Each time, the judgment function did not move. The firm grew. The juniors upgraded. The partners kept their pens. The pattern is the base rate you should use when you read the 2026 headline.
1975 · Typewriter to personal computer
The typewriter came for the typing pool. Partners kept their pens. The junior tier got laptops. The firm got bigger.
1995 · Fax to email
Email came for the fax room and overnight courier. Partners kept their pens. The firm moved faster on the same engagement shape.
2005 · Paper ledger to QuickBooks
QuickBooks came for the bookkeeper. Partners kept their pens. The CFO hired a younger team.
2015 · Cold call to HubSpot plus Calendly
CRM and scheduling came for the sales-coordinator role. Partners kept their pens. BD moved to content and inbound.
2026 · Back office to operator-layer AI
Clone comes for the invoicing tab, the transcript queue, the follow-up draft, and the dashboard refresh. Partners keep their pens. features.tsx says so.
The eight things no feature card touches
The half of the job Clone leaves alone on purpose.
None of the eight items below is automated by any card in features.tsx. Each is a thing clients hire a named human for. The operator layer sits below them, handling the paperwork so you have time to do them well.
What is still 100% you, after Clone is on your Mac
- The discovery call. Still you, still listening for the real problem.
- The scoping decision. Still your framework, your read of the client.
- The proposal price. Still your judgment on what advice is worth.
- The presentation of conclusions. Still your story in the room.
- The tough call on retention after a missed quarter. Still you.
- The referral to a lawyer, a tax accountant, a buyer. Still your network.
- The three sentences in an engagement letter that actually matter. Still you.
- The 'here is what I would do' at the end of the engagement. Still you.
Two sentences, two rituals
The Friday admin sweep, typed in plain English once.
This is a real Clone session, abbreviated. The second ritual deliberately never auto-sends: a discovery-call follow-up is a judgment artifact, and judgment artifacts stay on the human side of the line.
Replace the consultant vs augment the consultant
The two shapes of AI in consulting, side by side
Every row is one place the two postures disagree about where the line belongs.
| Feature | 'AI consultant' replacement pitch | Clone (operator layer) |
|---|---|---|
| What it does to discovery calls | 'AI consultant' products that pitch themselves as a replacement for human advice propose that an LLM can run the discovery call itself. The result is a generic synthesis; the client still wants a human to tell them what is load-bearing and what is not. | Clone does not sit on the call. It files the transcript, tags action items, and drafts your follow-up from your past emails. features.tsx line 32 through 37. You ran the discovery call. Clone made the notes filable. |
| What it does to proposal writing | A replacement product produces the whole proposal. The pricing is usually wrong because the LLM has not sat across from this client and seen the flinch. You then spend 45 minutes undoing it. | Clone drafts from your template, your prior proposals, and your pricing. The price is still your call. The three sentences that matter are still yours to write, with your margin for dignity. |
| What it does to framework selection | An 'AI does the consulting' product picks a framework from a library. The client pays a human $X0,000 per month because they want the judgment, not the framework library. | Clone has no framework library. Architecture principle 2 at line 52: 'your workflows, your voice.' The framework is yours. Clone executes the admin around it. |
| What it does to the recommendation | A replacement product prints a 'recommended action.' The client does not want a printed recommendation; they want a named professional to say the sentence in a room. | Clone prints nothing that reaches the client without you. The final recommendation is always yours. Clone wrote the cover page. |
| What it does to the back office | A replacement product tries to own everything. A VA at $3K to $6K per month tries to do everything a human can do about your admin. Both overshoot. | Clone takes exactly the ops work described in features.tsx and nothing else. No cap table, no client list, no pricing page decisions. comparison.tsx line 188: $49 per month vs $3K to $6K for a VA. |
| Where the data lives | Replacement platforms upload your client files to a cloud tenant. Your confidentiality clause did not plan for that. | Clone runs on your Mac. Architecture principle 1, line 47: 'client files, emails, contracts, and transcripts never leave your computer.' Local Llama is an option for Planner inference if you want full on-device. |
Under 10 minutes from worried headline to first ritual
Four steps. No OAuth carousel, no integrations page.
The first ritual fires on its next scheduled trigger. The Friday-3pm sweep fires next Friday. A post-Zoom-call ritual fires on your next call. There is no integrations marketplace to wait on.
pick a ritual → type one sentence → approve the first run → let it recur
Pick one ritual
The Friday admin sweep is the canonical starter. 45 minutes saved.
Type the sentence
Plain English. No connector marketplace, no trigger UI.
Approve the first run
Review the draft invoices, notes, and emails. Edit and send.
Let it recur
Next Friday at 3pm, the ritual fires again. You get a notification.
Want the admin-vs-judgment line drawn on your actual practice?
30 minutes with the Clone team. Bring your Friday admin list and we'll sketch two rituals you can ship on Monday.
Common questions from consultants reading the 2026 headline
Does 'AI is coming for consultants' mean consultants get replaced?
The honest answer is no, and the evidence is in how vendors in this category actually ship product. Clone's own feature file, src/components/features.tsx lines 13 through 61, has six cards. Every one automates back office work: invoicing, onboarding paperwork, transcript logging, follow-up drafting, dashboard assembly, or hours reclaimed. Zero cards mention discovery calls, framework choice, proposal pricing, or the recommendation itself. The vendor drew a line between admin and judgment, and built on one side of it. That is the practical shape of this change: the invoicing tab is automated; the judgment call is not.
How do you verify the 'zero cards mention strategy' claim?
Clone the cl0ne.ai repository. Open src/components/features.tsx. Read lines 13 through 61. Run 'grep -cE strategy|framework|discovery|recommendation|proposal|insight src/components/features.tsx.' The count is 0. Then run 'grep -nE title: src/components/features.tsx' and read the six titles: Invoicing on autopilot, Client onboarding in minutes, Zoom calls to CRM automatically, Follow-ups that feel personal, A dashboard you never had to build, Hours back every week. Those are the six. That is the full product surface the file decides to ship.
Which parts of a consulting engagement does Clone actually touch?
Five categories, all operational. One, invoicing: reads a time tracker, drafts in QuickBooks or FreshBooks, chases late payments you pre-approved. Two, onboarding admin: on a signed proposal, it provisions the shared workspace, drafts the kickoff agenda, books the call, sends the welcome email. Three, transcript logging: every call is transcribed by tl;dv, Fireflies, Otter, or native Zoom and filed on the right CRM contact. Four, follow-up drafting: it writes the email from the transcript plus your past tone, you still send. Five, dashboard assembly: a client-health board from Sheets plus CRM plus invoicing, refreshed daily. That is the complete list.
Which parts does Clone not touch?
The judgment layer. The discovery call itself, the framework and scoping call, the proposal pricing decision, the recommendation you say out loud at the end of the engagement, the retention conversation after a missed quarter, the hard referral to another professional, the three sentences in an engagement letter that actually matter. None of those appear as a feature card. Architecture principle 2 at line 52 is the vendor's own language for it: 'your workflows, your voice.' Clone observes and mirrors; it does not generate the conclusion.
Zapier, Make, and VAs all promise something like this. What is different here?
Three clean differences, all verifiable. One, configuration surface: Zapier and Make require you to draw triggers and branches in their UI. Clone takes one English sentence. Two, scope: HoneyBook and Dubsado replace your stack; Clone drives the stack you already pay for and leaves every SaaS subscription in place. Three, cost: the comparison row in src/components/comparison.tsx line 188 prints the numbers. Clone is $49 per month on Solo. A VA is $3K to $6K per month. The VA also takes weekends. Clone runs every Friday at 3pm whether you are in a hammock or not.
What is the shortest weekly ritual a worried consultant should pilot?
The Friday admin sweep. One sentence, saved once: 'every Friday at 3pm, reconcile this week's Timely hours, draft invoices for clients with more than four billable hours, write a HubSpot note on each, Slack me a digest.' The first fire is Friday, 3pm. You review the queue, approve, and get back to a client call. Typical reclaim: 45 to 50 minutes per Friday, zero new logins created, zero existing apps canceled. Once that ritual is stable, add a post-discovery-call follow-up drafter. Drafts only, never auto-send.
Does the 'AI on your laptop' model really keep NDAs intact?
On a properly configured install, yes. The Computer Agent and Memory are local by default. The Planner routes to a cloud LLM unless you route it to a local Llama, at which point no client file or transcript leaves your machine. Architecture principle 1, lines 46 through 49 verbatim: 'Clone operates your desktop apps from your desktop. Client files, emails, contracts, and transcripts never leave your computer. Your engagements stay confidential by default.' On an M-series Mac with 16 GB or more, local Planner inference is practical.
What if the AI is actually coming for the discovery call too?
Two things. First, it might eventually, for some tiers of the market, in a narrow window before clients push back on 'we need a named human in the room.' That is a general-purpose-AI question, not a Clone question. Second, the product bet visible in features.tsx is that the most defensible place to sit is the operator layer: admin automated, judgment retained. If the market moves to 'AI does the discovery call,' consultants who already reclaimed 10 to 15 hours per week on admin will be the ones with time to rebuild the relationship layer the new world will pay for.
Are there honest downsides to the operator-layer approach?
Three. First, if you have an app with no Mac session (for example, a client-issued managed VM), Clone cannot drive what you cannot open yourself. Second, ritual quality is only as good as the first sentence you write; some rituals need one iteration before they land in the queue the way you want. Third, a sufficiently old Mac with 8 GB of RAM is not practical for local Planner inference, so the Planner will route to the cloud; if your engagement rules out cloud inference, the upgrade to a newer Mac is the real purchase, not Clone itself.
How do I tell a worried junior consultant what this means for their career?
The pattern is old. In 1995 email came for the fax room. In 2005 QuickBooks came for the bookkeeping clerk. In 2015 Calendly and HubSpot came for the sales coordinator. In every case the firm kept its partners, sent more juniors to client meetings, and grew. The admin functions consolidated into a tool. The judgment and relationship functions did not. The 2026 version is the same shape: admin consolidates into an operator-layer AI, and the people who keep growing their practices are the ones who run the tool instead of competing with it.
How fast can a solo consultant get the first ritual running?
Under 10 minutes on a Mac. Install the .dmg, pick cloud or local inference, type one English sentence, approve the first run. The first ritual fires on its next trigger: a Friday-3pm ritual fires next Friday, a post-Zoom-call ritual fires on your next call. There is no connector page, no OAuth carousel, no integrations marketplace. comparison.tsx line 72 prints the 'set up in under 10 minutes' row that Zapier, HoneyBook, and a VA all get an X on.
Adjacent angles on the same line-drawing question
Related guides for a consultant evaluating AI in 2026
AI Tools for Consultants: Don't Grow the Stack, Operate the Stack
The sibling argument for the shopping-list keyword. Same anchor file, different framing: most guides hand you a 17 tool list; keep six tools and add one operator above them.
Best AI Tools for Independent Consultants 2026: The NDA-Safe Shortlist
Same product, different lens: the shortlist sorted by NDA compatibility. Read this one if your engagements routinely include a confidentiality clause.
AI Consultant for Small Business: The Operator You Hire, Not the App You Buy
The other end of the same trade: what clients mean when they say they want 'an AI consultant.' Usually they want a consultant with an operator, not a chatbot.