The ledger should track one more thing for agentic products: how much work the customer can trigger without asking you first. A custom workflow may look cheap at launch, then quietly multiply inference and review costs every month.
That changes the pricing conversation. Don’t just charge for building the exception. Set a usage boundary and a price for crossing it. Otherwise the biggest customer can take over your compute budget as well as your roadmap.
The pricing unit will shape what the team optimizes. Charge for tickets closed, and a fast closure can start to look like success even when the customer comes back tomorrow.
I’d test the metric with a shadow invoice before changing anyone’s bill. Show buyers what would have counted, what would not, and why. If that conversation reveals disputes, the definition needs work. If it reveals work the customer gladly pays for, you may have found your price.
The 13-week calendar is useful, but I would add one stress test: remove your largest expected payment and look at the lowest bank balance again. If a single late invoice takes you below payroll reserve, that customer is financing more of your business than you may realize.
I would also put an explicit gate between signing and starting customer-specific work. On a $24,000 annual deal with $3,000 of onboarding costs, don’t book the contractor because the contract is signed. Get the purchase order approved and collect an onboarding payment first. If procurement cannot move that quickly, move the start date. That is a calmer conversation than asking for faster payment after you have already spent the money.
The rule I’d keep beside the cash calendar is simple: a good deal should not require the rest of a tiny company to lend it money.
Dane, I’d apply that same test to revenue from a price increase. An extra $4,000 in MRR may appear to cover the hire, but it hasn’t survived a renewal cycle yet. Bank the uplift first. If customers stay, you can hire against revenue you’ve actually retained. If they don’t, you haven’t raised your buffer target by $48,000 on the strength of a good quarter.
A private fork sells more than engineering time. It gives one customer a claim on future release capacity, including when a security fix arrives at the worst possible moment. A maintenance fee based only on expected hours will underprice that promise.
I’d make the deal review ask one more question: which shared-product work might this obligation delay? If nobody can name the tradeoff, the fork is probably too cheap.
Priya, that’s the distinction: the audit trail shouldn’t just explain a charge after the fact. It should show whether a workflow is billable while there’s still time to fix a bad input or missing approval.
I’d make that status part of the product and the invoice preview. If a customer can spot and challenge a charge before month-end, pricing becomes easier to trust—and easier to renew.
Mara, yes. A separate renewal makes the obligation visible before it becomes unpaid work. I’d also spell out what happens if they don’t renew: support for their deployment ends, but an upstream merge doesn’t disappear. If we still expect four hours a month of upkeep, we need either a funded reserve or a feature we’d maintain for our other customers anyway.
That’s the decision I’d make before merging. Otherwise renewal protects the contract, not the two-person team’s calendar.
A maintenance fee makes the cost visible, but it can also make a fork look like a healthy product line. I’ve seen teams keep a divergence because the customer pays for it, even when nobody wants to own the next upgrade. Renewal needs a technical decision, not just a price review.
The open-versus-paid line is separate from the shared-versus-private line. You can charge for enterprise capabilities and still ship one tested release. That is usually a better promise to the customer and a safer one for the community.
A bridge can make a price change fairer. It can also hide a bad value metric. If customers doing the same work would pay substantially more, I’d question the unit before negotiating a longer allowance.
At renewal, the sharpest question may be: “What would you do differently if each additional document had a price?” If the answer is “use the product less,” the model may be charging for activity rather than value.
A usable exit path also makes pricing more honest. The alternative to your cloud is not “free software.” It is someone on the customer’s team owning upgrades, incidents, and recovery. That is the work a hosted service can charge to take away.
At renewal, the question becomes sharper: can the customer name what staying saves them this year? If not, making the door harder to use won’t fix the price.
The cash floor protects the business. I’d add a price for interruptions. A client who pays well but expects same-day responses can cost more product time than their hours on a timesheet suggest.
That makes response times part of the service offer, not just a personal boundary. Scheduled work gets one price. Immediate availability costs more. If a client won’t pay for the disruption, those protected product hours have a better use.
Eli, I’d draw one boundary carefully: an agent’s basic ability to stop, inspect, and undo a bad action should not depend on a paid tier. That’s part of making the free product safe to use. The commercial layer can fund the organizational work: centralized audit retention, approval policies, and someone accountable for recovery. For a small team, I’d name that boundary early, then ask two or three production users to pay for a pilot before building the controls. Otherwise “confidence” becomes another expensive feature list funded by the founder’s runway.
For an open-core product, the meter does more than set a price. It draws the line between what people can do freely and what they need to buy. I would keep self-hosted use genuinely free, then charge for the managed service and support my company actually provides.
I’d be careful with the advice to track every business outcome. In a customer-run deployment, you may not have the visibility to verify one. Requiring that telemetry just to make the pricing model work can cost more trust than the meter earns.
A shadow invoice tells you what customers would have paid if they had kept using the product the same way. That is a valuable billing test. It is not a demand forecast.
I’d put the largest increases in front of the affected customers before launch. Ask what they would change at that price. If the answer is “stop using the feature we bought you for,” the invoice may be correct, but the packaging is wrong.
Mara, that’s the test I’d put before any launch. Run the proposed meter against a customer’s past month of tickets and show them a sample invoice. If they can’t reconcile it with their own records, the definition of “resolved” needs work.
I’d also make reopenings and human handoffs visible in a running tally, not settle them as surprises at month-end. A meter only earns trust when the buyer can check it while there’s still time to act.
Dane, that’s the right pushback. Pricing a completed workflow aligns the bill with customer value, but it doesn’t cap what the vendor spends trying to complete it. A paid pilot should tell you what a job costs when it fails, not just when it works.
I’d make the stop rule part of the promise: “We’ll try up to this point. Then we’ll hand it off or ask before spending more.” That’s a limit customers can plan around, rather than one they discover when the product suddenly stops working.
Mara, “visible fence” is the right phrase. I’d make it visible to the customer and enforceable in the product. Show the remaining runs, warn before the limit, and pause before another job creates a cost you can’t recover. Then offer an upgrade paid in advance, not a surprise invoice afterward.
For a two-person shop, that’s a much smaller build than full hybrid billing. It protects cash flow without making a good customer feel punished for using the product.
For open-core agents, I’d draw the free/paid line at readiness, not memory. A self-hosted agent should be able to save its work and resume safely, even if that takes longer. Fast, dependable resumption is a reasonable managed-service promise to charge for.
If the free version forgets on purpose, the upgrade feels like a ransom. If the paid version passes an expired cache bill straight to the customer, it feels like a trap. Charge for a clear service level. Keep the cache mechanics—and their surprises—on your side of the invoice.
A meter can track value and still fail if the buyer can’t predict or verify the bill. “Issues resolved” sounds clean until a ticket reopens or a human steps in. Those rules need to be clear before the first invoice.
I’d test the proposed meter with one question: Can the customer estimate next month’s bill from data they already trust? If not, a platform fee doesn’t create predictability. It just puts a floor under the uncertainty.
Eli, that’s the right starting point. I’d make “accepted” explicit in the customer agreement, though. A draft someone clicks through is different from a support issue that stays resolved. Track acceptance now, but keep it separate from the business result. Otherwise a useful early signal can quietly become a billing dispute.
Priya, that’s a useful boundary: the customer can fund the work without owning the roadmap. I’d add a pricing test. Hosting and guarantees only become product revenue if the cost of delivering them stays predictable as customers grow. Track support hours and infrastructure cost per account, then price the managed offering around the reliability customers value. Otherwise, “open product, paid services” can still become an agency with a repository.
Maraya, “complexity should be earned” is the right rule. I’d keep an exception log before touching the billing system: who needed more, what it cost to serve them, and how long the manual fix took. Review it monthly. If three customers need the same exception and it is eating into margin, there may be a case for automating it. If not, keep handling it by hand. A few manual invoices are cheaper than a meter that takes a month to build and another hour every week to explain.
The paid-pilot test is useful, but founders building on open source need one more question: would this still be a product if the pilot customer stopped paying? A client can fund a feature without earning control of the roadmap.
I’d keep the reusable workflow in the open product. Charge for the work around it: hosting, migration, support, and guarantees. That line lets services finance development without turning every new customer into another custom project.
Mara, I’d take that question one step further: would the customer still pay the base fee in a month with little AI work? If not, “platform access” may describe our costs, but not their value. In that case, a base fee with included usage may feel more honest. The invoice should reflect what buyers think they bought, not just how we built it.
Maraya, the transition cost matters most when the contingency plan includes a price increase. New prices may improve margins, but they won’t fix this quarter’s cash gap. Customers need notice, annual contracts may delay collections, and a rushed change can create the churn you’re trying to survive. I’d model the reserve against cash you can collect under existing terms, then treat any pricing upside as a bonus.
This site uses cookies that are necessary for it to function (comments, logins, and saved preferences). We do not use advertising or third-party tracking cookies.
Eli Brandt on When Your Biggest Customer Starts Running Your Business
The ledger should track one more thing for agentic products: how much work the customer can trigger without asking you first. A custom workflow may look cheap at launch, then quietly multiply inference and review costs every month.
That changes the pricing conversation. Don’t just charge for building the exception. Set a usage boundary and a price for crossing it. Otherwise the biggest customer can take over your compute budget as well as your roadmap.
Maraya on Your AI Cost Unit Is Not Your Customer’s Value Unit
The pricing unit will shape what the team optimizes. Charge for tickets closed, and a fast closure can start to look like success even when the customer comes back tomorrow.
I’d test the metric with a shadow invoice before changing anyone’s bill. Show buyers what would have counted, what would not, and why. If that conversation reveals disputes, the definition needs work. If it reveals work the customer gladly pays for, you may have found your price.
Dane Whitlock on Revenue Isn’t Cash: A Bootstrapper’s Guide to Getting Paid on Time
The 13-week calendar is useful, but I would add one stress test: remove your largest expected payment and look at the lowest bank balance again. If a single late invoice takes you below payroll reserve, that customer is financing more of your business than you may realize.
I would also put an explicit gate between signing and starting customer-specific work. On a $24,000 annual deal with $3,000 of onboarding costs, don’t book the contractor because the contract is signed. Get the purchase order approved and collect an onboarding payment first. If procurement cannot move that quickly, move the start date. That is a calmer conversation than asking for faster payment after you have already spent the money.
The rule I’d keep beside the cash calendar is simple: a good deal should not require the rest of a tiny company to lend it money.
Mara Delgado on The Twelve-Month Cash Buffer: Why Bootstrappers Should Hoard Runway Instead of Chasing Growth
In reply to Dane Whitlock
Dane, I’d apply that same test to revenue from a price increase. An extra $4,000 in MRR may appear to cover the hire, but it hasn’t survived a renewal cycle yet. Bank the uplift first. If customers stay, you can hire against revenue you’ve actually retained. If they don’t, you haven’t raised your buffer target by $48,000 on the strength of a good quarter.
Eli Brandt on Your Enterprise Fork Is a Liability, Not a Feature
A private fork sells more than engineering time. It gives one customer a claim on future release capacity, including when a security fix arrives at the worst possible moment. A maintenance fee based only on expected hours will underprice that promise.
I’d make the deal review ask one more question: which shared-product work might this obligation delay? If nobody can name the tradeoff, the fork is probably too cheap.
Maraya on Credit Abstractions and the Illusion of Simplicity: Why Your Pricing Metric Is Lying to You
In reply to Priya Raman
Priya, that’s the distinction: the audit trail shouldn’t just explain a charge after the fact. It should show whether a workflow is billable while there’s still time to fix a bad input or missing approval.
I’d make that status part of the product and the invoice preview. If a customer can spot and challenge a charge before month-end, pricing becomes easier to trust—and easier to renew.
Dane Whitlock on When a Customer Pays for an Open-Source Feature
In reply to Mara Delgado
Mara, yes. A separate renewal makes the obligation visible before it becomes unpaid work. I’d also spell out what happens if they don’t renew: support for their deployment ends, but an upstream merge doesn’t disappear. If we still expect four hours a month of upkeep, we need either a funded reserve or a feature we’d maintain for our other customers anyway.
That’s the decision I’d make before merging. Otherwise renewal protects the contract, not the two-person team’s calendar.
Priya Raman on Your Enterprise Fork Is a Liability, Not a Feature
A maintenance fee makes the cost visible, but it can also make a fork look like a healthy product line. I’ve seen teams keep a divergence because the customer pays for it, even when nobody wants to own the next upgrade. Renewal needs a technical decision, not just a price review.
The open-versus-paid line is separate from the shared-versus-private line. You can charge for enterprise capabilities and still ship one tested release. That is usually a better promise to the customer and a safer one for the community.
Mara Delgado on The Renewal Is Where Your New Pricing Model Gets Tested
A bridge can make a price change fairer. It can also hide a bad value metric. If customers doing the same work would pay substantially more, I’d question the unit before negotiating a longer allowance.
At renewal, the sharpest question may be: “What would you do differently if each additional document had a price?” If the answer is “use the product less,” the model may be charging for activity rather than value.
Eli Brandt on Make It Easy to Leave Your Open-Source Cloud
A usable exit path also makes pricing more honest. The alternative to your cloud is not “free software.” It is someone on the customer’s team owning upgrades, incidents, and recovery. That is the work a hosted service can charge to take away.
At renewal, the question becomes sharper: can the customer name what staying saves them this year? If not, making the door harder to use won’t fix the price.
Maraya on From Client Work to SaaS Without a Runway Cliff
The cash floor protects the business. I’d add a price for interruptions. A client who pays well but expects same-day responses can cost more product time than their hours on a timesheet suggest.
That makes response times part of the service offer, not just a personal boundary. Scheduled work gets one price. Immediate availability costs more. If a client won’t pay for the disruption, those protected product hours have a better use.
Dane Whitlock on Don’t Sell Support. Sell Confidence.
In reply to Eli Brandt
Eli, I’d draw one boundary carefully: an agent’s basic ability to stop, inspect, and undo a bad action should not depend on a paid tier. That’s part of making the free product safe to use. The commercial layer can fund the organizational work: centralized audit retention, approval policies, and someone accountable for recovery. For a small team, I’d name that boundary early, then ask two or three production users to pay for a pilot before building the controls. Otherwise “confidence” becomes another expensive feature list funded by the founder’s runway.
Priya Raman on The Meter Is the Strategy: Choosing the Right Unit of Value for Your AI Product
For an open-core product, the meter does more than set a price. It draws the line between what people can do freely and what they need to buy. I would keep self-hosted use genuinely free, then charge for the managed service and support my company actually provides.
I’d be careful with the advice to track every business outcome. In a customer-run deployment, you may not have the visibility to verify one. Requiring that telemetry just to make the pricing model work can cost more trust than the meter earns.
Mara Delgado on Run a Shadow Invoice Before You Ship a New Price
A shadow invoice tells you what customers would have paid if they had kept using the product the same way. That is a valuable billing test. It is not a demand forecast.
I’d put the largest increases in front of the affected customers before launch. Ask what they would change at that price. If the answer is “stop using the feature we bought you for,” the invoice may be correct, but the packaging is wrong.
Maraya on The Meter Is the Strategy: Choosing the Right Unit of Value for Your AI Product
In reply to Mara Delgado
Mara, that’s the test I’d put before any launch. Run the proposed meter against a customer’s past month of tickets and show them a sample invoice. If they can’t reconcile it with their own records, the definition of “resolved” needs work.
I’d also make reopenings and human handoffs visible in a running tally, not settle them as surprises at month-end. A meter only earns trust when the buyer can check it while there’s still time to act.
Eli Brandt on The Casino Vibe: Why Rationed Limits Are the Tell of a Broken Pricing Model
In reply to Dane Whitlock
Dane, that’s the right pushback. Pricing a completed workflow aligns the bill with customer value, but it doesn’t cap what the vendor spends trying to complete it. A paid pilot should tell you what a job costs when it fails, not just when it works.
I’d make the stop rule part of the promise: “We’ll try up to this point. Then we’ll hand it off or ask before spending more.” That’s a limit customers can plan around, rather than one they discover when the product suddenly stops working.
Dane Whitlock on The Hybrid Trap: Why Most AI Pricing Strategies Collapse in the Middle
In reply to Mara Delgado
Mara, “visible fence” is the right phrase. I’d make it visible to the customer and enforceable in the product. Show the remaining runs, warn before the limit, and pause before another job creates a cost you can’t recover. Then offer an upgrade paid in advance, not a surprise invoice afterward.
For a two-person shop, that’s a much smaller build than full hybrid billing. It protects cash flow without making a good customer feel punished for using the product.
Priya Raman on Your Agent Went Idle. Your Inference Bill Didn’t.
For open-core agents, I’d draw the free/paid line at readiness, not memory. A self-hosted agent should be able to save its work and resume safely, even if that takes longer. Fast, dependable resumption is a reasonable managed-service promise to charge for.
If the free version forgets on purpose, the upgrade feels like a ransom. If the paid version passes an expired cache bill straight to the customer, it feels like a trap. Charge for a clear service level. Keep the cache mechanics—and their surprises—on your side of the invoice.
Mara Delgado on The Meter Is the Strategy: Choosing the Right Unit of Value for Your AI Product
A meter can track value and still fail if the buyer can’t predict or verify the bill. “Issues resolved” sounds clean until a ticket reopens or a human steps in. Those rules need to be clear before the first invoice.
I’d test the proposed meter with one question: Can the customer estimate next month’s bill from data they already trust? If not, a platform fee doesn’t create predictability. It just puts a floor under the uncertainty.
Maraya on The Meter Is the Strategy: Choosing the Right Unit of Value for Your AI Product
In reply to Eli Brandt
Eli, that’s the right starting point. I’d make “accepted” explicit in the customer agreement, though. A draft someone clicks through is different from a support issue that stays resolved. Track acceptance now, but keep it separate from the business result. Otherwise a useful early signal can quietly become a billing dispute.
Eli Brandt on From Client Work to SaaS Without a Runway Cliff
In reply to Priya Raman
Priya, that’s a useful boundary: the customer can fund the work without owning the roadmap. I’d add a pricing test. Hosting and guarantees only become product revenue if the cost of delivering them stays predictable as customers grow. Track support hours and infrastructure cost per account, then price the managed offering around the reliability customers value. Otherwise, “open product, paid services” can still become an agency with a repository.
Dane Whitlock on Your Pricing Page Needs a Segmentation System, Not Another Toggle
In reply to Maraya
Maraya, “complexity should be earned” is the right rule. I’d keep an exception log before touching the billing system: who needed more, what it cost to serve them, and how long the manual fix took. Review it monthly. If three customers need the same exception and it is eating into margin, there may be a case for automating it. If not, keep handling it by hand. A few manual invoices are cheaper than a meter that takes a month to build and another hour every week to explain.
Priya Raman on From Client Work to SaaS Without a Runway Cliff
The paid-pilot test is useful, but founders building on open source need one more question: would this still be a product if the pilot customer stopped paying? A client can fund a feature without earning control of the roadmap.
I’d keep the reusable workflow in the open product. Charge for the work around it: hosting, migration, support, and guarantees. That line lets services finance development without turning every new customer into another custom project.
Maraya on The Hybrid Trap: Why Most AI Pricing Strategies Collapse in the Middle
In reply to Mara Delgado
Mara, I’d take that question one step further: would the customer still pay the base fee in a month with little AI work? If not, “platform access” may describe our costs, but not their value. In that case, a base fee with included usage may feel more honest. The invoice should reflect what buyers think they bought, not just how we built it.
Mara Delgado on The Twelve-Month Cash Buffer: Why Bootstrappers Should Hoard Runway Instead of Chasing Growth
In reply to Maraya
Maraya, the transition cost matters most when the contingency plan includes a price increase. New prices may improve margins, but they won’t fix this quarter’s cash gap. Customers need notice, annual contracts may delay collections, and a rushed change can create the churn you’re trying to survive. I’d model the reserve against cash you can collect under existing terms, then treat any pricing upside as a bonus.