Agentic Development, Competition, and the Economics of Getting to First Revenue
Two founders start building the same small B2B SaaS. One delegates implementation tasks to coding agents and has a pilot-ready product in four weeks. The other takes seven. Both still need to find a manager with an urgent problem, earn approval to handle real customer data and get an invoice paid. Which founder has invested less when the first payment arrives?
In one fictional scenario below, the faster founder spends €6,000 less on development and €9,000 more getting to payment. The first receipt arrives a week later and costs €3,000 more to reach. In another scenario, the same build saving survives a smaller increase in commercial effort. The result depends on how much of the engineering saving remains after everything else that happens before cash arrives.
What the productivity research actually measures
Code completion suggests edits as a developer types. Interactive assistance lets a developer ask for explanations, patches or tests while directing the work. A coding agent can take a multi-step task, inspect the repository, modify files, run tools and revise its work before handing it back. The same interface can support more than one mode. Record the amount of work delegated and the human effort needed to accept the result.
- Cui et al. pooled three randomized field experiments at Microsoft, Accenture and an anonymous Fortune 100 company, covering 4,867 developers. The tool suggested code completions. The pooled estimate for tool use was 26.08% more completed tasks, with a standard error of 10.3 percentage points. Randomized access underpins the estimate, which concerns developers induced to use the tool. A change in output per period cannot simply be read as hours saved per task. Management Science, DOI 10.1287/mnsc.2025.00535
- METR's early-2025 experiment randomized access to AI at the task level for 16 experienced open-source developers working on 246 tasks in familiar, mature repositories. They mainly used Cursor Pro with Claude 3.5 or 3.7 Sonnet. Measured task duration increased by 19% with AI allowed, with a reported interval of 2% to 39%. The main measure was self-reported implementation time before and after pull-request review, with screen recordings available for an additional view of work patterns. It captures a mix of interactive and agentic features, without an isolated estimate for either mode. METR study paper
- METR's 24 February 2026 update reports preliminary changes in task time of −18% for returning developers (interval −38% to +9%) and −4% for new participants (−15% to +9%). Developers increasingly declined participation when asked to work without AI. Task selection and accounting for simultaneous work also complicated the comparison. METR regarded the new estimates as unreliable for current productivity and announced plans to change its study design. METR productivity update
- A 2026 Microsoft rollout study examined command-line coding agents and estimated roughly 24% more merged pull requests among adopters over its four-month window. It used an observational comparison and counted merged PRs as output. Release readiness and maintenance cost require different evidence. Murphy-Hill et al., preprint
- METR's 20 November 2026 capabilities report evaluates agent performance on tasks relevant to AI research and development. Models often matched or exceeded skilled human baselines on selected coding and optimization tasks. Where empirical feedback was weak, the report found indications of poorer judgment than expert humans. The combined TH1.1 and TH2 benchmark's fitted 50% time horizon doubled roughly every 3.6 months. That horizon is calibrated to how long a task takes a human and describes benchmark success probability. It does not measure the agent's runtime or tell a founder how quickly a product will ship. METR capability update
For a release decision, log active human minutes and elapsed calendar time separately. A founder can run an agent during customer interviews, but still spend the next morning reviewing the patch. Apply the same acceptance criteria in either workflow. For a small multi-tenant SaaS, check tenant isolation, backups and restoration, regression tests, dependency review and the operational handover. Track review and rework even when generation took seconds. A tool bill and a pile of accepted patches tell you little about the future cost of maintaining the result. Token Economics for a Solo Business
When cheaper entry changes what the buyer sees
Imagine a hosted service that reminds independent industrial equipment service firms about upcoming maintenance and exports a record of completed work. It imports assets from CSV, schedules reminders and records completion. The buyer is the operations manager at a firm with 10–30 technicians. Success might mean fewer missed service deadlines and less time assembling evidence, measured against the firm's recent records.
Lower build costs can bring in suppliers when several teams can reach the same buyers and deliver an acceptable replacement. The buyer has to consider their offers interchangeable enough to compare. Ten more generic reminder apps may have little effect if only one supports the firm's data import and export workflow. A new entrant can also serve a niche that previously had too little demand to justify the build.
If similar offers reach the same buyers, the following effects are plausible for the maintenance SaaS.
- Discovery and qualification. More suppliers bid for the same search terms or appear in the same trade directories. Broader promotion can bring more inquiries from firms with too few assets, no deadline problem or no budget owner. Track qualified accounts per channel and cost per qualified account alongside inquiry volume.
- Pricing and sales effort. A manager with three credible quotes can ask for a discount or a more specific proof of value. The founder may spend more time on a sample import, security answers and follow-up. A discount reduces receipts and contribution margin. Additional meetings increase the acquisition ledger.
- Time to payment. A broader shortlist can add demos and an approval round. An incumbent can include reminders in an existing contract, internal IT can build a scheduled spreadsheet workflow, and the operations manager can decide the current process is good enough. Record which alternative actually ends the deal. A slow internal approval and a lost head-to-head comparison call for different changes.
There are countereffects. One equipment distributor can introduce the founder to several qualified firms. Existing reach or a narrow segment can make a buyer easier to identify. A sample export that eliminates a weekly reconciliation step can give an operations manager a reason to move now. Agents may also reduce the human time spent researching accounts, preparing campaign variants or drafting import instructions. Verify account fit and claims, track tool and review costs, and leave approvals and customer commitments with the founder. Sales and Marketing for AI Software · One Human Owner, an AI Workforce
What market data can tell us
RevenueCat's 2026 report presents Appfigures data showing monthly new subscription app launches across iOS and Android rising from roughly 2,000 in January 2022 to more than 14,700 in January 2026. iOS represented about 77% by the report's latest comparison. The January 2026 revenue-by-launch-cohort chart also uses Appfigures: apps launched before 2020 accounted for 69% of the reported subscription revenue, and apps launched from 2025 onward accounted for 3%. Most other metrics in the report come from RevenueCat's own subscription dataset. The launch series describes mobile supply. Its timing overlaps with adoption of AI tools, and it does not identify what caused the increase or how many new apps were direct substitutes for a given buyer. RevenueCat, State of Subscription Apps 2026
The competition for attention extends across the wider software market. A maintenance SaaS can face only a few direct substitutes and still share search results, directories, ads and buyers’ inboxes with a growing supply of apps. As teams use coding agents to launch more products, that shared distribution space can get more crowded. In a crowded channel, reaching the same number of qualified buyers may now require more marketing effort or spend just to be noticed. For the maintenance SaaS, that could mean publishing a page around a specific service workflow, producing a demo with realistic sample data or paying to reach the right operations managers. Check whether it takes more founder hours or cash per qualified account in that channel. The mobile launch counts show the growth in supply. They do not measure the extra marketing spend a B2B founder needs.
There is also evidence closer to market entry. A September 2026 preprint by Cao and Bhatia found more first-time digital venture launches after late 2022 in product categories with greater estimated exposure to AI coding. Their research design compares categories using pre-LLM descriptions and links entry, traffic and financing. The exposure measure is a proxy. The study does not follow the competing products each buyer evaluated or their vendors' costs to win the first customer. Cao and Bhatia, preprint v3
In a separate B2B context, SaaS Capital's 2026 spending survey drew more than 1,000 private B2B SaaS responses. Median selling spend was 15% of ARR, versus 13% in the prior edition. Median marketing spend held at 8%. ARR is the denominator. Those ratios describe departmental spend at surveyed companies and cannot give euros per newly acquired customer or the budget to get a first payment. SaaS Capital, 2026 Spending Benchmarks
Repeated Ebsta/Pavilion reports cover opportunity win rates and sales cycles, and Ebsta separately studies qualification using CRM opportunities. ChartMogul's 2026 survey covers free-to-paid conversion for 200 B2B products. Benchmarkit's successive reports track acquisition efficiency using metrics such as sales and marketing expense per unit of new ARR and CAC payback. Those sources use different populations and funnel endpoints. In the public sources reviewed through 22 November 2026, I found no panel that tracks agent use, direct substitutes shortlisted by buyers, qualified leads, conversion to cash, sales-cycle length and first-customer acquisition costs for comparable new B2B SaaS products. The direction and magnitude of the combined effect remain open. Ebsta/Pavilion GTM reports · Ebsta qualification study · ChartMogul conversion report · Benchmarkit 2025 · Benchmarkit 2026
A fictional cost ledger from idea to first receipt
Each case starts with the same product scope, target buyer, acceptance criteria and founder rate of €75 per active hour. The 30-day pilot covers one team and one import. Every buyer gets the same €600 price and payment terms, with payment due before the pilot starts. Pre-payment onboarding includes sample data mapping and billing setup. All figures, dates and outcomes in the example are assumptions. Amounts exclude VAT.
Start the clock when work on the idea begins. Stop on the first cleared customer payment, or at the week-12 cutoff if no money arrives. Show cash paid to suppliers separately from the imputed cost of founder time. The latter values the founder's hours for comparison and does not imply a salary was paid. The €600 receipt is reported separately from the gross cost of reaching it. Track later refunds as their own events.
On a narrow screen, scroll each wide table to see the remaining columns.
| Scenario | Pilot-ready | First payment | Cash paid | Founder time | Total |
|---|---|---|---|---|---|
| Baseline | Week 7 | Week 10 | €1,750 | 310 h / €23,250 | €25,000 |
| Agentic · lower | Week 4 | Week 8 | €3,000 | 240 h / €18,000 | €21,000 |
| Agentic · equal | Week 4 | Week 9 | €3,400 | 288 h / €21,600 | €25,000 |
| Agentic · higher | Week 4 | Week 11 | €4,300 | 316 h / €23,700 | €28,000 |
| Agentic · unpaid | Week 4 | None by week 12 | €4,200 | 304 h / €22,800 | €27,000 |
The agentic cases use the same build budget. They vary the route to a paying customer. The less expensive route still uses more sales effort and marketing spend than the baseline, but a partner intro curbs the extra work. In the equal-cost case, the founder spends more time explaining the offer and pursuing stalled leads. In the higher-cost case, comparisons and approvals add more hours, campaign costs and hosting before payment. All agentic cases reach a pilot-ready release in week 4. The baseline reaches it in week 7. In the equal-cost case, payment arrives earlier at the same total cost. In the higher-cost case, an earlier release still leads to a later receipt.
Hours and paid expenses per activity are shown as hours / cash. Subtotals include both.
| Activity | Baseline | Agentic · lower | Agentic · equal | Agentic · higher | Agentic · unpaid |
|---|---|---|---|---|---|
| Implementation and architecture | 160 h / €0 | 72 h / €0 | 72 h / €0 | 72 h / €0 | 72 h / €0 |
| Review, tests and security validation | 25 h / €500 | 31 h / €500 | 31 h / €500 | 31 h / €500 | 31 h / €500 |
| Rework and maintainability | 25 h / €0 | 23 h / €0 | 23 h / €0 | 23 h / €0 | 23 h / €0 |
| Release engineering and operational readiness | 20 h / €250 | 18 h / €250 | 18 h / €250 | 18 h / €250 | 18 h / €250 |
| Coding agents | 0 h / €0 | 0 h / €450 | 0 h / €450 | 0 h / €450 | 0 h / €450 |
| Customer interviews | 12 h / €0 | 12 h / €0 | 12 h / €0 | 12 h / €0 | 12 h / €0 |
| Positioning and marketing | 18 h / €750 | 22 h / €1,350 | 34 h / €1,550 | 40 h / €2,250 | 40 h / €2,250 |
| Demos, follow-up and commercial review | 16 h / €0 | 19 h / €0 | 32 h / €0 | 39 h / €0 | 30 h / €0 |
| Unsuccessful and still-open prospecting | 20 h / €0 | 26 h / €0 | 46 h / €0 | 56 h / €0 | 60 h / €0 |
| Hosting and routine operations | 6 h / €250 | 7 h / €450 | 10 h / €650 | 13 h / €850 | 14 h / €750 |
| Onboarding and billing before cutoff | 8 h / €0 | 10 h / €0 | 10 h / €0 | 12 h / €0 | 4 h / €0 |
| Build subtotal | €18,000 | €12,000 | €12,000 | €12,000 | €12,000 |
| Non-build subtotal | €7,000 | €9,000 | €13,000 | €16,000 | €15,000 |
Each ledger entry belongs to one activity. Interviews are problem discovery. Marketing includes campaign research, positioning and distribution. Sales time for the account that eventually pays goes in demos and follow-up. The time spent on accounts that have yet to pay, whether lost or still open, goes in the prospecting row. In the cutoff case the demos and commercial review row covers the signed but unpaid account. The hosting row covers ongoing infrastructure and routine operations. Release engineering covers one-time work to ship and operate safely. If a coding subscription also supports sales work, allocate the invoice across those activities once.
How much extra effort can the build saving fund?
The change in total cost equals the increase in non-build cost minus the net development saving. The baseline build costs €18,000. The agent-assisted build costs €12,000 after including the agent bill and review. The development saving is €6,000, made up of 86 hours saved × €75 minus €450 extra tool spend. The rest of the baseline costs €7,000. The break-even budget for everything outside the build is therefore €13,000.
In the equal-cost case, sales rises by €3,150, positioning and marketing by €2,000 and the remaining costs by €850. They use the full €6,000. If marketing and the remaining costs are already €2,850 above baseline, sales can rise by at most €3,150, equivalent to 42 more hours at €75 if it involves only founder time.
Hold the non-build increase at €6,000 and vary development independently. If extra review and rework reduce the build saving to €4,500, first payment costs €26,500, or €1,500 more than baseline. If the build saves €7,500 at the same quality bar, the total falls to €23,500, or €1,500 less. For cash runway, compare cash columns directly. The equal-cost case ties on economic cost and requires €1,650 more cash.
At week 12 in the unpaid case, the founder has spent €4,200 cash and 304 hours worth €22,800, for €27,000 incurred. The pilot agreement is signed and the invoice unpaid. Record that outcome as open at cutoff. Its eventual cost and payment date remain unknown. With zero acquired customers, CAC is undefined.
Test willingness to pay before adding scope
Start with interviews about the last missed service, the cost of catching it late, how it was handled and who could approve a change. Ask comparable qualified firms to buy the same €600 pilot on the same terms. Show a narrowly scoped demo using sample data and agree on a success criterion, such as time to produce a maintenance record compared with their existing process. Where the product is still incomplete, state the delivery date and any manual work included before asking the buyer to commit.
Track stages with dates. An interested reply establishes contact. A qualified account has the target workflow, a named budget owner and a reason to act within a defined period. Test access records evaluation. A signature records agreement. An invoice is a receivable. A cleared payment establishes that an account paid under those terms. In this example, count it as retained when the account completes the paid pilot and pays its first recurring invoice. For a direct monthly subscription, use the first paid renewal. A single receipt shows that one customer paid once.
Measure CAC and payback after the first sale
Cost to first payment includes building the product and exploring the market. For a later sales cohort, customer acquisition cost (CAC) measures the sales and marketing spending allocated to that cohort divided by new paying accounts it produced. Include failed attempts and any acquisition labor at the same rate. Define the attribution window in advance, and show open deals separately so a recently started cohort can mature. Keep initial product development out of this cohort numerator.
In a separate fictional cohort, €2,400 cash and 128 founder hours at €75 produce ten new customers who have paid their first €350 subscription invoice. Acquisition costs €12,000 in total, or €1,200 per new payer. Cash-only CAC is €240. Suppose each active account costs €35 per month in variable hosting and payment fees and €65 in support, including founder time. Monthly contribution is €250 per active account. Simple contribution-based CAC payback is €1,200 ÷ €250 = 4.8 active months at that contribution, or five full billed months at monthly granularity.
If two of the ten cancel after month two and eight renew through month six, the cohort contributes €5,000 during the first two months and €2,000 each month thereafter. The cumulative contribution is €11,000 after month five and €13,000 after month six, so the cohort recovers its €12,000 acquisition cost during month six. That still leaves shared overhead, ongoing product maintenance and fixed costs to cover. Use actual payments, credits, refunds and support time to update the calculation as renewals arrive.
A measurement template for your next launch
Keep three small logs. Link each cost item to a single category and each commercial event to an account ID. Use the accounts you actually targeted as the denominator, and keep channel and cohort dates consistent when comparing rates.
| Log | Minimum fields |
|---|---|
| Cost ledger | Date, activity, category, account or shared cohort, active founder hours, hourly rate, cash paid, invoice or time-entry ID |
| Funnel events | Account ID, segment, channel, date, reached / interested / qualified / demo or test / signed / invoiced / paid / renewed, shortlisted alternatives, lost reason or open status |
| Receipts | Account ID, invoice ID, due date, settlement date, net amount, payment or refund, pilot / recurring / renewal |
Set a review date before starting. Sum cash and the value of founder time through the first settlement or cutoff. Record release and payment dates, lead-to-qualified and qualified-to-paid conversion by cohort, and days from qualification to signature and signature to cash. Use buyer-confirmed alternatives and recurring objections to decide whether to improve the product, narrow the audience or try a different channel. The €6,000 threshold in the example gives the founder up to €6,000 of extra room in the non-build budget. It still takes a paying buyer to establish whether any of that budget was well spent.
When a cheaper build costs more overall
A growing supply of software can make it more expensive to be seen, including for products with few direct substitutes. For a founder relying on crowded channels, reaching the same number of qualified buyers may take more marketing spend or founder time. Longer sales cycles can add demos, follow-up and operating costs. If these extra costs exceed the net development saving, the founder spends more before the first payment, even after an earlier launch. In the fictional case, €9,000 in extra non-build costs outweighs €6,000 in build savings, raising the total from €25,000 to €28,000.
The evidence reviewed supports this as a plausible risk. It does not establish that total costs are already rising across B2B SaaS or that increases are more common than decreases. Where the founder can reach buyers through a focused channel, referrals or an existing audience, more of the development saving can survive to first payment.