Party School · Go deeper · Stack design

Build your stack on purpose.

You already own a stack. The question is no longer which six tools exist, it's whether the ones you pay for are priced right for the way you actually use them. This lesson is the economics: tools by role, then cost per outcome, then switching cost. Where a flat seat beats a metered call, where it doesn't, and the exact volume the two lines cross.

Updated August 1, 2026 Refreshed monthly Sources: Anthropic · OpenAI · Google · Perplexity · Midjourney · Gamma · MCP
Core concept 01

A stack is a set of roles you pay to fill, not a shelf of apps.

Read every tool through three questions, in order. What role does it play. What does one outcome cost on it. What would it cost to leave. Role, then cost per outcome, then switching cost. That order is the whole model, and it is deliberate: sorting by price first is how people end up with six half-used subscriptions, because everything looks affordable at $20. Sort by role first and the flood collapses into three slots. A hub you open all day, one generalist chosen for depth. A bench of specialists you rent by the job. And an agent layer that runs unattended, which is where the pricing model quietly changes underneath you.

Price is the second pass. And the honest price of a tool is never its sticker, it is its sticker divided by the outcomes you actually produce on it. A $20 seat you touch three times a month costs almost $7 an outcome. The same $20 spread across two hundred outcomes costs a dime. Two people pay the same invoice and run wildly different businesses. The tier table from the plain-English version of this lesson still holds and still moves monthly; check it before you pay. The job here is the decision math that sits on top of it.

Go deeper: the third axis is the one vendors hope you skip

Role and cost per outcome are visible on the pricing page. Switching cost is not, which is exactly why it belongs in the model. It has three parts you can measure: how cleanly your data exports, how far your prompts travel to another model, and what it costs to swap the model underneath a workflow you have already tuned. A tool can win on role and cost per outcome and still be the wrong buy because leaving it later means rebuilding, not re-tuning.

Hold the three axes together and most tool decisions answer themselves. A cheap seat with high lock-in on a task you will run for years is more expensive than it looks. A metered call with zero lock-in on a task you run twice a week is cheaper than it looks.

Sources: Anthropic · OpenAI · Google · Perplexity · Midjourney · Gamma · Model Context Protocol

Scroll the diagram sideways →

Sort by role first, then price each role ROLE 01 · HUB One generalist Claude · ChatGPT · Gemini pick one, go deep FLAT SEAT · ~$20/mo Cheaper per outcome the more you run it. Human in the loop all day. ROLE 02 · SPECIALIST A rented bench Perplexity · Midjourney · Gamma swap freely by job SEAT, ON WHILE IN USE Pay only the months the job is live. Low lock-in, so no home to defend. ROLE 03 · AGENT / API Runs unattended Cowork · agent mode · MCP software calls the model METERED · $/OUTCOME Near zero when idle, rises per run. Scales past a human's hours. same models underneath · three different cost curves
Three roles, three cost models · price each role on its own curve
The stack-decision grid · role × cost model × lock-in · confirm every price on the vendor's page
Role in your stackDefault cost modelFlip to a metered API / MCP call whenLock-in risk
Hub · daily generalistFlat seatKeep itAlmost never. You are human-in-the-loop all day; the seat is already cheapest per outcome.High: memory, projects, muscle memory. Keep a portable brief.
Research · sourced answersSeat, burstyThe lookups are repeatable and you run them below your crossover volume, or want them unattended.Low: outputs are text and citations you keep.
Images · on-brand assetsSeat while shippingYou batch dozens on a schedule and a metered image call undercuts the tier.Low to medium: styles port, saved refs may not.
Decks · client-facingSeat, seasonalRarely worth metering; low volume, human taste in the loop.Medium: export to slides before you cancel.
High-volume draftingMeteredConvertThe task is templated, repeated, and automatable. This is the classic seat-that-should-be-a-call.Low if the prompt is a file you own, not app memory.
Core concept 02

A seat is a fixed cost. A call is a marginal cost. They meet the business differently.

A subscription does not move. It is the same invoice whether you produce one outcome this month or three hundred, which means a flat seat rewards volume: every additional run drags the average cost per outcome down toward zero. A metered API or MCP call is the opposite shape. It is close to nothing when idle and adds a few cents on every run, so it rewards restraint: you pay for exactly what you use and nothing for the weeks you don't. Neither is cheaper in general. Each is cheaper in a range.

The number that decides it is cost per outcome, and the honest version divides the seat price by the outcomes you actually produced, not the ones you pictured when you signed up. Most people never run that division, which is why a stack drifts. Two seats you barely touch quietly cost more per result than one you lean on hard, and the metered alternative for the light ones is often pennies. The worksheet below runs the division for one task, both ways, and finds the volume where the two ways of paying cost exactly the same.

Go deeper: why the metered line ignores two real costs

The clean crossover math treats a metered call as pure per-token spend. Reality adds two lines the chart leaves out. First, build cost: someone has to wire the call, handle errors, and keep it running, which a seat gives you for free as a finished app. Second, the app's convenience: a UI, memory, and no maintenance have real value even when the raw tokens are cheaper elsewhere.

So the true break-even sits to the right of the mathematical one. You do not flip a task to an API the instant the token cost dips under the seat; you flip when the volume, times the per-outcome gap, clears the build cost and buys back something the app never could, like running at 3am unattended.

Sources: Anthropic API docs · OpenAI API pricing · Model Context Protocol

The worksheet you leave with

Cost per outcome, both ways, and where they cross

Pick one recurring outcome and run it twice: once as a share of a flat seat, once as a metered call. The example numbers are illustrative for 2026; swap in your own, and confirm today's per-token rate on the model's API pricing page before you trust the last line.

  1. 1Name one recurring outcome.e.g. a sourced research brief, a product listing, a client recap. One repeatable result.
  2. 2Flat seat price that produces it, per month.example: $20/mo. The subscription you already pay for this job.
  3. 3Outcomes you actually produce a month.the honest count, not the aspirational one. example: 15.
  4. 4Flat cost per outcome = line 2 ÷ line 3.$20 ÷ 15 ≈ $1.33 each. That is what one result really costs on the seat.
  5. 5Metered cost per outcome = tokens × API rate.example: ~35k tokens in and out at an illustrative blended rate ≈ $0.17. Check the API pricing page for the real number.
  6. 6Crossover volume = line 2 ÷ line 5.$20 ÷ $0.17 ≈ 118 outcomes/mo. Below it the seat costs more than metered; above it the seat is cheaper.
  7. Which side of the line is this task on?below crossover AND automatable → candidate to become an API/MCP call. Above it, or human-driven all day → keep the seat.
Core concept 03

The crossover runs both directions. Cross it on purpose.

Plot the two costs and the decision draws itself. A flat seat is a horizontal line: same price at any volume. A metered call is a sloped line from zero, rising a few cents per run. They meet at one point, the crossover, which is just the seat price divided by the per-outcome rate. Left of it, at low volume, metered is the cheaper line. Right of it, at high volume, the flat seat wins and it isn't close. Everything about the buy follows from which side of that point a given task lives on.

That gives you two clean triggers to move. A subscription becomes an API or MCP call when a task is automatable, repeatable, and either sitting far left of its crossover on an idle seat or needs to run unattended, which no seat can do. It goes the other way, a metered call becomes a subscription, when you find yourself paying by the token for something a human drives interactively all day at high volume, where a flat seat caps the bill and hands you a finished interface. The mistake is never crossing. It is drifting: paying seat prices for machine work, or metering work a person does by hand.

Source: Anthropic, "Building Effective Agents"

Scroll the diagram sideways →

Where a seat and a metered call cost the same $ / MO OUTCOMES / MO → CROSSOVER ≈ 118/mo Flat seat · fixed ~$20/mo Metered · $/outcome LOW VOLUME metered is cheaper → convert to a call HIGH VOLUME seat is cheaper → keep the subscription The real flip sits right of the crossover: add build cost and the app's convenience before you move.
The subscription-to-API crossover · one point decides the direction
Core concept 04

Switching cost is the third number, and build-versus-buy is where you spend it.

Lock-in has three measurable parts, and none of them are your monthly fee. Data export: how cleanly your history, files, and outputs come out in a format you can reuse, versus staying trapped in one app. Prompt portability: how far a prompt you tuned travels to a different model before it needs rework, because a prompt shaped around one model's quirks rarely lands the same on another. And model-swap cost: what it takes to change the brain underneath a workflow you have already built around. A tool can be cheap per outcome and still be expensive to own, because leaving it later means rebuilding instead of re-tuning.

This is where the buy-versus-build line actually sits. Buy, meaning keep the finished app and its seat, until an outcome is high-volume, automatable, and the vendor's margin has become your largest variable cost. Build, meaning move it to a metered API or MCP call you assemble, once it clears that bar and you want it running unattended. The single move that keeps every one of these doors open costs nothing: hold your standing brief and your best examples in a plain document you own, not inside one app's memory. Do that and switching drops from rebuilding to re-tuning, and the commitment to a hub stays a choice you can revisit instead of a cage.

Go deeper: budget for the model swap you can't see coming

The models under your stack get retired and replaced on the vendor's schedule, not yours. A prompt tuned hard against one version can drift when the version changes, even inside the same app, which is a lock-in cost that arrives without you switching anything. The defense is the same practice engineers use for code: keep a small set of evals, a handful of real inputs with the output you expect, and re-run them whenever the model moves.

That turns a model swap from a mystery into a measurement. If the evals still pass, the swap is free. If they don't, you know exactly what to re-tune before it reaches a client, and you learn it on your bench instead of in production.

Scroll the diagram sideways →

The portability spectrum how much of your setup walks out the door with you PORTABLE · WALKS WITH YOU LOCKED IN · STAYS BEHIND Plain-text prompt Exported data + brief Saved projects, custom GPTs Proprietary memory Fine-tune, embedded flow The rule: enjoy the locked-in features, but keep the portable copy of your setup outside the app.
Lock-in spectrum · commit to a hub, keep the portable version outside it
Fresh from the lab

What changed this month

The agent-layer column just got two new entries, both still seat-priced. On July 9, OpenAI launched ChatGPT Work, which takes a goal instead of a prompt, gathers context across your connected apps, and runs independently for hours to hand back a finished artifact. It rolled out on Pro, Enterprise, and Edu first, with Plus and Business to follow, alongside a ChatGPT for Small Business program bundling partner trials. Notably, it ships inside the flat seat, not as a metered add-on, so for now the role-03 column stays priced like role-01: a subscription, just one that runs longer per invocation.

Claude Cowork's session model moved server-side, which is the more interesting change than the platform expansion. On July 7, Anthropic brought Cowork to web and mobile, but the real architecture shift is that sessions now execute on Anthropic's infrastructure instead of your device, so a scheduled task survives a closed laptop or a locked phone. That is what "runs unattended" actually requires, compute that outlives the client. It's in beta on Claude Max first, still inside the flat-seat price, no per-run metering disclosed yet.

Midjourney shipped a model upgrade with zero pricing change. Version 8.2 became the default on July 24, with Midjourney citing better aesthetics and a Personalization system that reads your rated history more accurately. Every tier from Basic up inherits it automatically, a clean example of a vendor absorbing a capability jump into an existing seat rather than gating it behind a new tier.

Sources: OpenAI, "ChatGPT is now a partner for your most ambitious work", July 9, 2026 · OpenAI, "Introducing the ChatGPT for small business program" · Claude by Anthropic, "Claude Cowork on web and mobile", July 7, 2026 · Midjourney, "Version"

Vocabulary

Six words for pricing a stack

You know what the tools do. These are the words for how they charge, which is the part that decides your stack once the feature list stops being the interesting question.

API vs app

Two doors to the same model. An app is a human at a UI, priced per seat. An API is software calling the model directly, priced per token. Same brain underneath, opposite cost curve.

Tokens / metered pricing

A token is the unit models bill in, roughly three-quarters of a word. Metered means you pay per token in and out, so cost scales straight with volume instead of sitting flat.

Cost per outcome

Total spend on a tool divided by the finished results it produced. The one number that compares cleanly across a flat seat and a metered call, and the one almost nobody actually computes.

Lock-in

How much of your setup and data stays trapped when you leave. Low if your brief is a text file you own; high if it lives in proprietary memory or a fine-tune only that vendor runs.

MCP

The Model Context Protocol, an open standard that lets a model call your tools and data directly. It's the plumbing that turns a chat app into an agent, and a subscription task into a metered call.

Self-host vs hosted

Hosted: the vendor runs the model and you rent access. Self-host: you run an open-weights model on your own hardware, no per-token fee but real operating cost. Only pays at serious scale.

Deeper definitions, straight from the builders: Anthropic Academy · MCP docs · OpenAI Academy

Make it yours

Audit your stack by role and cost per outcome

Pick your lane. Each panel runs the same advanced pass on a real stack: inventory the tools by role, compute cost per outcome on the heaviest recurring task, and find the one seat that should become a metered call or the one tool to cut. Not a starter kit. A pruning.

For female founders

I'm the CEO, the marketing team, and the bookkeeper before 9am, which means the actual building waits for everyone else to log off.

Two seats, one metered candidate

Audit: Claude Pro ($20/mo) hub + Perplexity Pro ($20/mo) specialist
  1. Inventory by role: Claude is the hub you drive all day, Perplexity is a specialist you open in bursts. Two flat seats, $40/mo, different usage shapes.
  2. Cost per outcome: count the sourced scans you actually ran last month. If it's a handful, Perplexity is costing you several dollars a scan for work an automatable research call does for cents.
  3. Spot the flip: the recurring competitive scan is templated and schedulable, the textbook seat-that-should-be-a-call. Keep Claude as the hub, and put the scan behind a metered call once it runs weekly.
  4. Portability: keep your positioning, ICP, and voice in a document you own, so the hub is a commitment you can revisit, not a cage.
  5. 30-day unit-cost review: recompute cost per outcome on both seats; drop or convert the one that isn't earning its flat fee.

The payoff: one deliberate hub, one metered specialist, and a founder paying for volume she uses instead of two seats she half-touches.

For artists

I make the work, and then grant applications and artist statements quietly eat the studio hours the work actually needed.

The seasonal seat is the leak

Audit: Gemini AI Pro ($19.99/mo) hub + Perplexity Pro ($20/mo) specialist
  1. Inventory by role: Gemini writes statements and applications, Perplexity finds and vets open calls. One you use most weeks, one you use in application season only.
  2. Cost per outcome: if you chase grants eight times a year, a year-round Perplexity seat is roughly $30 per search. That is a seat priced for volume you don't have.
  3. Spot the flip: drop Perplexity to free or a metered search between cycles, and pay only the months a deadline is live. Nothing here is worth a standing subscription off-season.
  4. Portability: keep your CV, voice, and a plain-language paragraph about your practice as a file, so switching hubs is a paste, not a rebuild.
  5. 30-day unit-cost review: tie it to your next deadline, not the calendar date, and cut any seat that sat idle through a full off-season.

The payoff: the research tool only bills when a deadline is real, and the studio stops subsidizing a seat it doesn't use.

For actors

Between the survival job and the self-tapes, the paperwork of the career crowds out the actual acting.

Meter the lookups, keep the hub

Audit: ChatGPT Plus ($20/mo) hub + Perplexity Pro ($20/mo) specialist
  1. Inventory by role: ChatGPT drafts cover notes and slate scripts, Perplexity does the pre-submission lookup on the casting director and the project. One daily, one occasional.
  2. Cost per outcome: count real submissions per month. A pre-submission lookup is small and repeatable, so a full research seat is expensive per use unless you're submitting weekly.
  3. Spot the flip: flex Perplexity to free between audition seasons; the lookup is the automatable one, cheap as a metered call the weeks it isn't a daily habit.
  4. Portability: keep your resume, three bio lengths, and your tone note in a document, not locked inside one custom GPT.
  5. 30-day unit-cost review: review what you booked against what each tool touched, and keep only the seat that's carrying its weight.

The payoff: the hub earns its seat daily, and the research bills like the burst it actually is.

For musicians

Nobody hired me to be my own manager, but I'm the one booking gigs, chasing deposits, and answering wedding inquiries at midnight.

Don't pay year-round for a twice-a-year deck

Audit: ChatGPT Plus ($20/mo) hub + Gamma Plus ($10/mo) specialist
  1. Inventory by role: ChatGPT handles pitch emails, captions, and inquiry replies daily, Gamma builds the EPK and one-sheet you touch a couple times a year.
  2. Cost per outcome: a Gamma seat paid every month for two EPK refreshes a year costs about $60 per deck. The seat is priced for a cadence you don't run.
  3. Spot the flip: keep Gamma free between release cycles and pay a single month when you rebuild. ChatGPT stays paid because inquiry replies are a real daily load.
  4. Portability: keep your bio, past press, and the five details fans always ask in a file, so any deck tool can start from it.
  5. 30-day unit-cost review: after a release cycle, note what got used, and downgrade the seat that idled between drops.

The payoff: the deck tool bills per release, not per month, and the daily hub is the only standing subscription.

For fitness pros

By the time class lets out I still owe posts, DMs, and the renewal texts to three people who went quiet.

Batch the graphics before you upgrade the tier

Audit: Claude Pro ($20/mo) hub + Midjourney Basic ($10/mo) specialist
  1. Inventory by role: Claude runs programming, check-ins, and content drafts as the daily hub; Midjourney makes challenge graphics and flyers in bursts.
  2. Cost per outcome: divide the Midjourney tier by the images you actually export a month. Occasional use keeps Basic's cost-per-image low; a higher tier only pays if you batch heavily.
  3. Spot the flip: if graphics become a monthly batch, price a metered image call against the tier before you upgrade. Programming stays on the seat, since it's human-in-the-loop weekly.
  4. Portability: keep your method, movement standards, and check-in style in a Project brief you can export, not trapped in app memory.
  5. 30-day unit-cost review: Sunday programming day is your built-in review; recompute cost per image and hold the lower tier until volume forces the move.

The payoff: you pay the tier your volume justifies, not the one the upsell suggests, and the studio still looks sharp.

For coaches

I sell transformation, and then my actual week disappears into scheduling, notes, and the follow-ups I keep meaning to send.

Pay the deck seat only in enrollment months

Audit: Gemini AI Pro ($19.99/mo) hub + Gamma Plus ($10/mo) specialist
  1. Inventory by role: Gemini writes recaps and follow-ups all week; Gamma builds the program overview deck new clients ask for, which is enrollment-cycle work, not weekly.
  2. Cost per outcome: a year-round Gamma seat for two enrollment pushes a year is a high cost per deck. Weekly writing on Gemini, by contrast, drives its seat cost per outcome down fast.
  3. Spot the flip: keep Gamma paid only the months you're actively enrolling; free the rest. The recap and follow-up load keeps Gemini on all year.
  4. Portability: hold your niche, method, and two proven pieces of content in a file, so the hub choice stays reversible.
  5. 30-day unit-cost review: tie it to your enrollment cycle so the review lands when you can still act on the number.

The payoff: the deck seat tracks your enrollment calendar, and the hub is the one thing paid every month because it works every week.

For therapists

I trained to help people, and somewhere along the way I became the office manager too.

The red line first, then trim the idle seat

Audit: Claude Pro ($20/mo) hub + Gamma Plus ($10/mo) specialist
  1. Draw the line first: no client information goes into any tool, seat or API, ever. Clinical notes need a HIPAA-compliant tool with a signed BAA, a different aisle entirely.
  2. Inventory by role: Claude handles FAQ replies, waitlist emails, and directory drafts; Gamma builds the referral one-pager and any CEU slides, which is low-volume work.
  3. Cost per outcome: the referral kit is a few outcomes a year, so a standing Gamma seat is expensive per deck. Claude's practice-admin load justifies its seat.
  4. Spot the flip: flex Gamma to free between projects and pay only the month you build; there is nothing here worth metering, since none of it is high-volume or automatable.
  5. 30-day unit-cost review: confirm the red line held and recompute whether either seat earned its month.

The payoff: the privacy line stays intact, and you pay the deck seat only when a deck exists.

For authors

I wanted to write the next book. Instead I run launches, newsletters, and reader email like an accidental small publisher.

The newsletter is the automatable one

Audit: Gemini AI Pro ($19.99/mo) hub + Perplexity Pro ($20/mo) specialist
  1. Inventory by role: Gemini writes newsletters, synopsis variants, and event copy; Perplexity finds podcasts, book clubs, and comp titles, which is launch-window work.
  2. Cost per outcome: the newsletter is weekly and formulaic, so it drives real volume; the pitch research is bursty, so a year-round research seat is expensive per search off-season.
  3. Spot the flip: the newsletter is the templated, repeated task that's a candidate to become a metered draft. Let Perplexity lapse to free between books; keep the hub for the weekly writing.
  4. Portability: keep your synopsis, a sample chapter, reviews, and your reader voice in a document, so the automatable draft can run against any model.
  5. 30-day unit-cost review: count replies, not sends, and cut the seat the numbers don't justify.

The payoff: platform upkeep drops to a metered draft plus an edit, and the research seat only bills in launch season.

For chefs

Menus, allergy notes, and quotes all live in my head and my texts until the day something finally slips.

Structured quotes want to be a call

Audit: Gemini AI Pro ($19.99/mo) hub + Midjourney Basic ($10/mo) specialist
  1. Inventory by role: Gemini writes menus, catering quotes, and specials posts; Midjourney makes seasonal menu cards and flyer art in bursts.
  2. Cost per outcome: quotes are your highest-volume repeatable task and follow a fixed shape, dishes times per-head cost. Menu art is occasional, so Basic's cost per image stays low.
  3. Spot the flip: once your dishes, sourcing, and per-head costs are fixed inputs, the quote is templated enough to become a metered call. Midjourney Basic stays as-is for the light art load.
  4. Portability: keep signature dishes, sourcing philosophy, and price points in a file, so the quote logic isn't trapped in one app.
  5. 30-day unit-cost review: after the season turns, check whether the quote task is cheaper as a call than as a seat you drive by hand.

The payoff: the after-midnight quote shift becomes a call you trigger in seconds, and the menu still sounds like you.

For consultants

Billable work eats the whole day, so my own pipeline, proposals, and invoices always get pushed to tomorrow.

Meter the pre-call research, keep the delivery seat

Audit: Claude Pro ($20/mo) hub + Perplexity Pro ($20/mo) specialist
  1. Inventory by role: Claude drafts proposals, recaps, and thought-leadership; Perplexity runs pre-call research on the prospect and their market, which is per-deal and bursty.
  2. Cost per outcome: if deals arrive less than weekly, an always-on research seat is expensive per brief. Proposals and recaps, by contrast, are steady work that justifies the hub.
  3. Spot the flip: the pre-call research is templated and burstable, a candidate for a metered call you fire per deal. Let Perplexity lapse during delivery-heavy stretches.
  4. Portability: keep your positioning, the three problems you solve, and a winning proposal in a file, so the delivery brief travels.
  5. 30-day unit-cost review: your Friday pipeline review is the natural moment to ask whether each seat earned its month.

The payoff: research bills per deal, not per month, and the pipeline stays warm while you're billing.

For health advocates

Every case is a crisis and the paperwork never stops, so growing the practice always loses to today's emergency.

Templates are one-time builds, not standing seats

Audit: ChatGPT Plus ($20/mo) hub + Gamma Plus ($10/mo) specialist
  1. Set the privacy rule first: no client health information in any tool or API, ever. Work in templates and blanks; the person's details get added by you, after, in your own documents.
  2. Inventory by role: ChatGPT builds appeal and intake templates with [BRACKETS]; Gamma makes workshop and community-education decks, both low-volume once built.
  3. Cost per outcome: a template is a one-time build, so paying a seat to regenerate what you already have is waste. The seat's real job shrinks once the library exists.
  4. Spot the flip: flex Gamma to free between workshops; keep ChatGPT only as long as the template library is still growing, then reassess the seat.
  5. 30-day unit-cost review: check whether the library actually grew, and drop the seat that stopped producing new outcomes.

The payoff: the paperwork mountain becomes a library you built once, not a subscription you keep renewing to reprint.

For makers

I make it, list it, ship it, answer every message about it, and somehow run out of time to actually make more of it.

Listings are the batch job that wants a call

Audit: ChatGPT Plus ($20/mo) hub + Midjourney Standard ($30/mo) specialist
  1. Inventory by role: ChatGPT writes listings, wholesale pitches, and market applications; Midjourney mocks up seasonal collections, which is drop-driven, not year-round.
  2. Cost per outcome: listings are per-SKU and templated, your highest-volume repeatable task. Midjourney Standard's fast hours only earn out around drops, so it's overpriced between them.
  3. Spot the flip: the per-SKU listing is a candidate to become a metered batch call once the format is fixed. Drop Midjourney to Basic or pause it between collections.
  4. Portability: keep your materials, process, price ranges, and three best listings as a voice file, so the batch call runs against any model.
  5. 30-day unit-cost review: line it up with your restock cadence and recompute cost per listing and per mockup.

The payoff: listing night becomes a batch call, and the image tier flexes with the drops instead of billing flat all year.

For nonprofit leaders

I'm the grant writer, the thank-you-note writer, and the program, all at once, and the mission is stuck behind my inbox.

Boilerplate wants a call, funder search is seasonal

Audit: ChatGPT Plus ($20/mo) hub + Perplexity Pro ($20/mo) specialist
  1. Inventory by role: ChatGPT drafts grant boilerplate and donor communications; Perplexity finds funders whose guidelines match your mission, which is prospecting-season work.
  2. Cost per outcome: LOI boilerplate is repeated and templated, so it carries real volume; the funder search is bursty, so a year-round research seat is expensive per cycle.
  3. Spot the flip: the boilerplate LOI is a candidate to become a metered draft once your template is set. Keep Perplexity paid only during active prospecting, free between cycles.
  4. Portability: keep your mission, programs, impact numbers, and one funded proposal in a file, so the draft call has a stable source.
  5. 30-day unit-cost review: tie it to your grant calendar so the review doubles as deadline planning.

The payoff: boilerplate drafts itself as a call, the research seat bills only in season, and funders still hear a human.

For photographers

I shoot weekends, edit all week, and the inquiries and gallery follow-ups fall through whatever cracks are left.

The same-day reply is the future call

Audit: Claude Pro ($20/mo) hub + Gamma Plus ($10/mo) specialist
  1. Inventory by role: Claude drafts same-day inquiry and gallery-delivery emails daily; Gamma builds the pricing and package guide, a one-time build with seasonal updates.
  2. Cost per outcome: inquiry replies are daily and human-reviewed, so the hub seat is well priced. The pricing guide is a few outcomes a year, so its seat should flex.
  3. Spot the flip: the same-day inquiry reply is repeatable and automatable, a candidate for a metered responder later. Keep Gamma paid only while the guide is being updated.
  4. Portability: keep your packages, pricing, turnaround, and three strong replies as a file, so the responder can run against any model.
  5. 30-day unit-cost review: compare inquiry-to-booking rate against tool spend, and trim the seat that isn't moving it.

The payoff: same-day replies without a VA, and a pricing-guide seat that bills only when the guide changes.

For realtors

My old clients are a graveyard in my CRM, and every new listing turns into a scramble of copy and coordination.

Listing copy is the batch, nurture is the human part

Audit: Gemini AI Pro ($19.99/mo) hub + Midjourney Basic ($10/mo) specialist
  1. Inventory by role: Gemini writes listing descriptions and nurture emails; Midjourney makes just-listed and open-house graphics in bursts around new inventory.
  2. Cost per outcome: listing descriptions are per-property and templated, high-volume in season; nurture emails are lower-volume and voice-driven. Midjourney Basic covers occasional art.
  3. Spot the flip: the per-listing description is a candidate to become a metered batch call in a heavy season; nurture stays human. Upgrade the image tier only during a listing surge.
  4. Portability: keep your farm area, niche, recent sales, and client voice in a file, so the batch call has a stable brief.
  5. 30-day unit-cost review: a monthly planning hour is the natural moment to recompute cost per listing and per graphic.

The payoff: listing copy batches as a call in season, nurture stays personal, and the graveyard turns into a referral engine.

For stylists

I'm behind the chair all day, so rebooking texts and content pile up, and the clients I forget just don't come back.

Rebooking texts are the schedulable call

Audit: ChatGPT Plus ($20/mo) hub + Midjourney Basic ($10/mo) specialist
  1. Inventory by role: ChatGPT writes rebooking nudges and captions between clients; Midjourney makes mood boards you show before you touch a strand, a light and occasional load.
  2. Cost per outcome: rebooking texts are high-volume, repeated, and templated, your most automatable task. Mood boards are a couple a week, so Basic's cost per image stays comfortable.
  3. Spot the flip: the rebooking nudge is the candidate to become a scheduled, metered send off your rebooking window. Midjourney Basic stays as-is for the light art load.
  4. Portability: keep your services, prices, rebooking windows, and the products you believe in as a file, so the send logic isn't trapped in a custom GPT.
  5. 30-day unit-cost review: your day off is the natural review, since it's the thing this leaner stack is trying to protect.

The payoff: rebooking runs as a scheduled call instead of an evening of texts, and the chair stays booked.

The house rule

Lauren's line, at engineering altitude: what belongs to you and keeps your humanity, you keep. What frees you to be more human away from your laptop, you hand off. The deeper version adds a second clause: hand it off to the cheapest shape that does the job well. A seat for the work a person drives all day, a metered call for the work a machine should run unattended, and a portable brief so no vendor owns the version of you that made either one work.

Next lesson: Automate the boring stuff →