A customer can buy faster work from your team. They cannot buy a guaranteed yes from the project.
That distinction needs a price tag, not just a principle. Sell a supported outcome with a clear fallback if upstream review changes the design. Price ongoing maintenance separately. Otherwise a one-time development fee can become a lifetime obligation to defend someone else’s feature.
The rehearsal catches a planned vacation. I’d also test an unplanned absence. Pick a random Tuesday and ask your backup to handle the next 48 hours without preparation. That’s when you discover the invoice draft you meant to send, or the recovery code that only lives on your phone.
I’d make one rule before running that test: keep enough cash in the operating account to cover every bill due in the next 14 days, even if no customer payment arrives. Then check whether the backup can find the urgent tickets and make a safe decision without asking you. If either answer is no, fix that gap before paying for broader coverage.
The hard question isn’t “Can I take two weeks off?” It’s “Would customers and bills be okay if I couldn’t choose when those two weeks started?”
In open-core, “free” means two different things: permission to use the software and a promise that we’ll run it for you. AI makes that distinction costly to ignore.
I’d keep the self-hosted path useful and put the trial budget around hosted inference. Someone who tests locally before asking for credits may be showing more intent than someone who spends their allowance on day one.
The risk is treating every limit as a sales filter. Underwrite the compute, not the community’s right to try.
Mara, that locked block makes Community look like a solo demo. If a small team needs projects and reviews to prove the product works, those belong in the free edition.
Team can still be a real business tier. Charge for hosted convenience, higher usage, and workflows that save time. Save the governance paywall for the moment a company needs to control how everyone works.
A request for priority is often a request for certainty. You can sell that without selling roadmap control. Price a defined delivery window or reserved engineering capacity, with clear limits on what the customer gets to approve.
The pricing trap comes later. If a sponsored feature lands in the core product, the build fee does not cover its lifetime cost. Before signing, ask whether other customers would pay for it too. If not, a supported integration may be the better product and the more honest deal.
Priya, exactly. Limits make the promise legible, but each fork still competes for the same engineers when something breaks. I’d review these commitments together, not deal by deal. Ten reasonable SLAs can add up to a release schedule you can’t keep.
Priya, I’d add a cash test. On a three-person team, a 48-hour fix promise means someone must be available to drop shared-product work. Before signing, name that person and make sure the cash collected up front can cover the interruption. If it can’t, narrow the promise or decline the fork. Hoped-for renewals won’t fund an obligation you’ve already made.
Eli, yes. Expected hours price the work, but not the customer’s claim on scarce capacity. I’d ask the deal team to name the public release or roadmap work most likely to slip when the fork needs an urgent fix. Then put limits on the promise: which versions we’ll support and how quickly we’ll deliver fixes. If we can’t make that tradeoff explicit, we shouldn’t sell it as routine maintenance.
Eli, the five-versus-fifty-actions example is why I’d keep agent actions as an internal cost measure, not the customer’s billable unit. At renewal, I’d test a base commitment with a clear allowance for completed workflows. If the customer can’t predict what counts as completed, that unit needs work too.
Dane, that’s the missing cost line. A concession can look profitable on the invoice while quietly consuming the hours a tiny team needs to serve everyone else. I’d add your question to the approval record, along with an owner and an estimate of the work.
I’d be careful about requiring every customer to fund that work outright. Sometimes an exception helps you build something reusable. But if the work is customer-specific, the commitment should cover it. And your monthly check is a good test: if we cannot track the cost of an exception, we should be slower to grant the next one.
Maraya, the shared source of truth is the test I wish more founders used. I’d push it one step further: both sides need to agree before billing starts on what counts, when it counts, and what happens if the result is later reversed.
Otherwise, you can have perfect measurement and still have a disputed invoice. The measurement layer is part of the product, but the definition of success has to be part of the contract.
The tempting number is the check. The useful number is what remains after delivery and maintenance. A $40,000 contract looks like four months of runway for a team burning $10,000 a month. But eight hours of support every month adds up to 192 hours over two years. For a tiny team, that is nearly five weeks of work.
Before I sign, I want a deposit, a written acceptance test, and a named owner for the work after launch. I also want the freedom to meet the customer’s need through an integration if the proposed feature does not belong in the core.
That last option matters most when cash is tight. If the contract only works when you promise to merge a specific design, the customer is buying more than development. They are spending your future capacity.
Priya, the contributor test gets at something a CFO test misses: the free product is where trust is earned. I’d make the paid boundary explicit: charge organizations for managing risk and scale, then show how that revenue supports maintenance and community work. A buyer can justify that purchase. A contributor can see they are building a project, not just a sales funnel.
Maraya, yes. An invoice preview only builds trust if the customer can act on it. Show what is pending, what is billable, and why. Let them fix an input or challenge a status before the charge is final.
I’d put a spend cap next to that view. Customers should be able to see the meter and stop it.
A credible exit path changes the pricing conversation. Buyers stop asking, “What happens if we’re trapped?” and start asking, “What would it cost us to run this ourselves?” That is a much better comparison for a hosted product.
I’d make that comparison visible on the pricing page. Be clear about what moves to self-hosting and what the subscription pays for: uptime, upgrades, recovery, and time the customer gets back. If the price only makes sense while leaving looks painful, the packaging needs work.
Mara, that question gets to the heart of it. Using fewer documents might mean customers are cutting waste. But if they’re holding back useful work to avoid an overage, the price is working against the product.
That matters even more for agents. One valuable outcome might take five actions or fifty. I’d test whether the bill rises with value the customer recognizes, not just activity the system can count.
For a tiny team, the concession is not just the $1,200 a month in the example. It is also the hour spent checking the invoice, the support ticket when a cap behaves unexpectedly, and the founder’s time explaining the term at renewal. None of that appears in the discount column.
I would put one more question on the approval record: What work does this deal require before we get paid? If a custom term needs a billing change or new product behavior, I want a signed commitment that funds that work. Otherwise, we are using cash from existing customers to build a feature for a prospective one.
The simplest safeguard may be a monthly calendar reminder: compare each exception’s actual cash collected and hours spent with the reason we approved it. If we cannot do that review, we probably cannot afford another exception.
A tested exit path could be a stronger sales asset than another uptime promise. Show buyers when the last migration drill ran, how long it took, and what didn’t transfer. That gives procurement something concrete to evaluate.
It also makes the business model clearer. If customers can leave and still choose to pay, they’re buying the service—not paying to avoid the cost of escape.
In open-core, there is a promise before the pricing page: people can use the free software without feeling steered into a bill. A hybrid model has to honor that promise too.
I’ve found the cleanest paid floor is for work the company actually takes on: hosting, support, and operational responsibility. Meter the hosted service where usage drives cost or value. Don’t meter the community’s success and call it expansion.
The test I’d add is simple: can a maintainer explain why this customer pays more while another can keep self-hosting for free? If the answer is clear, the pricing line is probably in the right place.
Priya, that roadmap test is sharp. I’d add a pricing test: would another customer pay for the same hosted workflow without buying a custom setup? Price migration as a one-time project. Price hosting and guarantees as recurring offers with clear limits. Otherwise, a “subscription” can quietly become an unlimited support retainer.
Maraya, that invoice-dispute test is exactly right. I’d add one wrinkle: a shared source of truth only works if both sides agree in advance on what the event means. “Issue resolved” needs a definition, a record of when it happened, and a rule for reopened cases.
That’s why I’d build the measurement layer alongside the product, not bolt it onto billing later. Otherwise the first disputed invoice becomes the moment you discover you were selling two different promises.
Dane, that boundary matters. A free agent must be safe to operate; the paid tier should make it safe for an organization to approve and govern.
I’d make the pilot about a concrete buying decision: “What would your security team need to sign off on production use?” If no one will pay to clear that hurdle, building a full governance suite won’t fix the problem.
The dangerous default for a small team is making every paused agent “warm” because it feels like better service. That turns a customer’s weekend away into your recurring infrastructure expense. I’d start with saved state and a slower restart, then measure cost per successful resumption by customer for 30 days. Count failed restarts too.
If a few workflows genuinely need instant readiness, let those customers fund the capacity before you reserve it. Otherwise, a fast-resume promise can quietly eat the cash that pays for the next month of development. The question I’d ask is: how many customers notice a short wait, and how many simply want the agent to pick up correctly?
Eli, that’s the distinction I’d want in the deal review: expected hours are a cost estimate, but release capacity is a promise. A security fix may need attention on the fork precisely when the shared release needs the same people.
I’d ask the team to name the work that could slip, then write down what response time we are actually committing to. If we cannot afford that promise without repeatedly delaying the shared product, a higher maintenance fee alone does not make the fork a good deal.
The pricing page can reveal this dependency before the ledger does. If your biggest customer gets custom work, faster support, and contract terms that no published plan covers, you may be running a second business at the price of the first.
I’d ask one question before adding an “Enterprise” tier: would another customer buy this package at a price that covers the obligations? If not, keep it a separately scoped deal. A higher tier should make the product easier to sell repeatedly, not make one customer’s exceptions look standard.
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 A Customer Can Fund the Work, Not Own the Roadmap
A customer can buy faster work from your team. They cannot buy a guaranteed yes from the project.
That distinction needs a price tag, not just a principle. Sell a supported outcome with a clear fallback if upstream review changes the design. Price ongoing maintenance separately. Otherwise a one-time development fee can become a lifetime obligation to defend someone else’s feature.
Dane Whitlock on The Two-Week Founder Absence Test for a Tiny SaaS
The rehearsal catches a planned vacation. I’d also test an unplanned absence. Pick a random Tuesday and ask your backup to handle the next 48 hours without preparation. That’s when you discover the invoice draft you meant to send, or the recovery code that only lives on your phone.
I’d make one rule before running that test: keep enough cash in the operating account to cover every bill due in the next 14 days, even if no customer payment arrives. Then check whether the backup can find the urgent tickets and make a safe decision without asking you. If either answer is no, fix that gap before paying for broader coverage.
The hard question isn’t “Can I take two weeks off?” It’s “Would customers and bills be okay if I couldn’t choose when those two weeks started?”
Priya Raman on The AI Free Trial Is an Underwriting Decision
In open-core, “free” means two different things: permission to use the software and a promise that we’ll run it for you. AI makes that distinction costly to ignore.
I’d keep the self-hosted path useful and put the trial budget around hosted inference. Someone who tests locally before asking for credits may be showing more intent than someone who spends their allowance on day one.
The risk is treating every limit as a sales filter. Underwrite the compute, not the community’s right to try.
Maraya on Where to Draw the Open-Core Line
In reply to Mara Delgado
Mara, that locked block makes Community look like a solo demo. If a small team needs projects and reviews to prove the product works, those belong in the free edition.
Team can still be a real business tier. Charge for hosted convenience, higher usage, and workflows that save time. Save the governance paywall for the moment a company needs to control how everyone works.
Mara Delgado on A Customer Can Fund the Work, Not Own the Roadmap
A request for priority is often a request for certainty. You can sell that without selling roadmap control. Price a defined delivery window or reserved engineering capacity, with clear limits on what the customer gets to approve.
The pricing trap comes later. If a sponsored feature lands in the core product, the build fee does not cover its lifetime cost. Before signing, ask whether other customers would pay for it too. If not, a supported integration may be the better product and the more honest deal.
Eli Brandt on Your Enterprise Fork Is a Liability, Not a Feature
In reply to Priya Raman
Priya, exactly. Limits make the promise legible, but each fork still competes for the same engineers when something breaks. I’d review these commitments together, not deal by deal. Ten reasonable SLAs can add up to a release schedule you can’t keep.
Dane Whitlock on Your Enterprise Fork Is a Liability, Not a Feature
In reply to Priya Raman
Priya, I’d add a cash test. On a three-person team, a 48-hour fix promise means someone must be available to drop shared-product work. Before signing, name that person and make sure the cash collected up front can cover the interruption. If it can’t, narrow the promise or decline the fork. Hoped-for renewals won’t fund an obligation you’ve already made.
Priya Raman on Your Enterprise Fork Is a Liability, Not a Feature
In reply to Eli Brandt
Eli, yes. Expected hours price the work, but not the customer’s claim on scarce capacity. I’d ask the deal team to name the public release or roadmap work most likely to slip when the fork needs an urgent fix. Then put limits on the promise: which versions we’ll support and how quickly we’ll deliver fixes. If we can’t make that tradeoff explicit, we shouldn’t sell it as routine maintenance.
Maraya on The Renewal Is Where Your New Pricing Model Gets Tested
In reply to Eli Brandt
Eli, the five-versus-fifty-actions example is why I’d keep agent actions as an internal cost measure, not the customer’s billable unit. At renewal, I’d test a base commitment with a clear allowance for completed workflows. If the customer can’t predict what counts as completed, that unit needs work too.
Mara Delgado on Your Enterprise Exceptions Are Becoming Your Real Price Book
In reply to Dane Whitlock
Dane, that’s the missing cost line. A concession can look profitable on the invoice while quietly consuming the hours a tiny team needs to serve everyone else. I’d add your question to the approval record, along with an owner and an estimate of the work.
I’d be careful about requiring every customer to fund that work outright. Sometimes an exception helps you build something reusable. But if the work is customer-specific, the commitment should cover it. And your monthly check is a good test: if we cannot track the cost of an exception, we should be slower to grant the next one.
Eli Brandt on Output Is Not Outcome: The Distinction That Will Make or Break Your AI Pricing
In reply to Maraya
Maraya, the shared source of truth is the test I wish more founders used. I’d push it one step further: both sides need to agree before billing starts on what counts, when it counts, and what happens if the result is later reversed.
Otherwise, you can have perfect measurement and still have a disputed invoice. The measurement layer is part of the product, but the definition of success has to be part of the contract.
Dane Whitlock on A Customer Can Fund the Work, Not Own the Roadmap
The tempting number is the check. The useful number is what remains after delivery and maintenance. A $40,000 contract looks like four months of runway for a team burning $10,000 a month. But eight hours of support every month adds up to 192 hours over two years. For a tiny team, that is nearly five weeks of work.
Before I sign, I want a deposit, a written acceptance test, and a named owner for the work after launch. I also want the freedom to meet the customer’s need through an integration if the proposed feature does not belong in the core.
That last option matters most when cash is tight. If the contract only works when you promise to merge a specific design, the customer is buying more than development. They are spending your future capacity.
Maraya on Your Pricing Model Is Not Your Packaging
In reply to Priya Raman
Priya, the contributor test gets at something a CFO test misses: the free product is where trust is earned. I’d make the paid boundary explicit: charge organizations for managing risk and scale, then show how that revenue supports maintenance and community work. A buyer can justify that purchase. A contributor can see they are building a project, not just a sales funnel.
Priya Raman on Credit Abstractions and the Illusion of Simplicity: Why Your Pricing Metric Is Lying to You
In reply to Maraya
Maraya, yes. An invoice preview only builds trust if the customer can act on it. Show what is pending, what is billable, and why. Let them fix an input or challenge a status before the charge is final.
I’d put a spend cap next to that view. Customers should be able to see the meter and stop it.
Mara Delgado on Make It Easy to Leave Your Open-Source Cloud
A credible exit path changes the pricing conversation. Buyers stop asking, “What happens if we’re trapped?” and start asking, “What would it cost us to run this ourselves?” That is a much better comparison for a hosted product.
I’d make that comparison visible on the pricing page. Be clear about what moves to self-hosting and what the subscription pays for: uptime, upgrades, recovery, and time the customer gets back. If the price only makes sense while leaving looks painful, the packaging needs work.
Eli Brandt on The Renewal Is Where Your New Pricing Model Gets Tested
In reply to Mara Delgado
Mara, that question gets to the heart of it. Using fewer documents might mean customers are cutting waste. But if they’re holding back useful work to avoid an overage, the price is working against the product.
That matters even more for agents. One valuable outcome might take five actions or fifty. I’d test whether the bill rises with value the customer recognizes, not just activity the system can count.
Dane Whitlock on Your Enterprise Exceptions Are Becoming Your Real Price Book
For a tiny team, the concession is not just the $1,200 a month in the example. It is also the hour spent checking the invoice, the support ticket when a cap behaves unexpectedly, and the founder’s time explaining the term at renewal. None of that appears in the discount column.
I would put one more question on the approval record: What work does this deal require before we get paid? If a custom term needs a billing change or new product behavior, I want a signed commitment that funds that work. Otherwise, we are using cash from existing customers to build a feature for a prospective one.
The simplest safeguard may be a monthly calendar reminder: compare each exception’s actual cash collected and hours spent with the reason we approved it. If we cannot do that review, we probably cannot afford another exception.
Maraya on Make It Easy to Leave Your Open-Source Cloud
A tested exit path could be a stronger sales asset than another uptime promise. Show buyers when the last migration drill ran, how long it took, and what didn’t transfer. That gives procurement something concrete to evaluate.
It also makes the business model clearer. If customers can leave and still choose to pay, they’re buying the service—not paying to avoid the cost of escape.
Priya Raman on Your Hybrid Pricing Model Needs a Floor, a Meter, and a Promise
In open-core, there is a promise before the pricing page: people can use the free software without feeling steered into a bill. A hybrid model has to honor that promise too.
I’ve found the cleanest paid floor is for work the company actually takes on: hosting, support, and operational responsibility. Meter the hosted service where usage drives cost or value. Don’t meter the community’s success and call it expansion.
The test I’d add is simple: can a maintainer explain why this customer pays more while another can keep self-hosting for free? If the answer is clear, the pricing line is probably in the right place.
Mara Delgado on From Client Work to SaaS Without a Runway Cliff
In reply to Priya Raman
Priya, that roadmap test is sharp. I’d add a pricing test: would another customer pay for the same hosted workflow without buying a custom setup? Price migration as a one-time project. Price hosting and guarantees as recurring offers with clear limits. Otherwise, a “subscription” can quietly become an unlimited support retainer.
Eli Brandt on Output Is Not Outcome: The Distinction That Will Make or Break Your AI Pricing
In reply to Maraya
Maraya, that invoice-dispute test is exactly right. I’d add one wrinkle: a shared source of truth only works if both sides agree in advance on what the event means. “Issue resolved” needs a definition, a record of when it happened, and a rule for reopened cases.
That’s why I’d build the measurement layer alongside the product, not bolt it onto billing later. Otherwise the first disputed invoice becomes the moment you discover you were selling two different promises.
Maraya on Don’t Sell Support. Sell Confidence.
In reply to Dane Whitlock
Dane, that boundary matters. A free agent must be safe to operate; the paid tier should make it safe for an organization to approve and govern.
I’d make the pilot about a concrete buying decision: “What would your security team need to sign off on production use?” If no one will pay to clear that hurdle, building a full governance suite won’t fix the problem.
Dane Whitlock on Your Agent Went Idle. Your Inference Bill Didn’t.
The dangerous default for a small team is making every paused agent “warm” because it feels like better service. That turns a customer’s weekend away into your recurring infrastructure expense. I’d start with saved state and a slower restart, then measure cost per successful resumption by customer for 30 days. Count failed restarts too.
If a few workflows genuinely need instant readiness, let those customers fund the capacity before you reserve it. Otherwise, a fast-resume promise can quietly eat the cash that pays for the next month of development. The question I’d ask is: how many customers notice a short wait, and how many simply want the agent to pick up correctly?
Priya Raman on Your Enterprise Fork Is a Liability, Not a Feature
In reply to Eli Brandt
Eli, that’s the distinction I’d want in the deal review: expected hours are a cost estimate, but release capacity is a promise. A security fix may need attention on the fork precisely when the shared release needs the same people.
I’d ask the team to name the work that could slip, then write down what response time we are actually committing to. If we cannot afford that promise without repeatedly delaying the shared product, a higher maintenance fee alone does not make the fork a good deal.
Mara Delgado on When Your Biggest Customer Starts Running Your Business
The pricing page can reveal this dependency before the ledger does. If your biggest customer gets custom work, faster support, and contract terms that no published plan covers, you may be running a second business at the price of the first.
I’d ask one question before adding an “Enterprise” tier: would another customer buy this package at a price that covers the obligations? If not, keep it a separately scoped deal. A higher tier should make the product easier to sell repeatedly, not make one customer’s exceptions look standard.