You have an idea for a software business. You know the customers, understand the problem and can see why somebody would pay to have it solved. Then you start listing what needs to be built: a website, accounts, payments, the product itself, customer dashboards, an administration system. Before long, you’re planning a department.
That calculation has changed.
A focused SaaS company can now begin with one or two people doing work that would previously have been divided across developers, a designer, a technical lead and somebody responsible for infrastructure. AI agents provide much of the implementation capacity. The people provide the direction, judgment and responsibility.
For the founders I work with, that can be a straightforward partnership: you bring your understanding of the market, and I help turn it into a clear product vision, then build the software through Blueprint Ventures. You don’t need to recruit an engineering team before you can find out whether the business works.
In more than 25 years of building software, I haven’t seen a better time to do this.
Small beginnings, substantial businesses
Chatbase, TypingMind and Photo AI each began with a founder turning a new technical opportunity into something people would pay for. Their founders’ accounts show how a small beginning can lead to substantial revenue and usage.
Chatbase reached a reported $1 million in annual recurring revenue in 117 days. Yasser Elsaid launched it as a solo project in 2023, initially letting people upload documents and ask questions about them. In his Starter Story interview, he describes launching to sixteen Twitter followers and repeatedly demonstrating the product as he added features. A useful product and a clear demonstration gave him a way to reach customers from a tiny starting audience.
Chatbase subsequently grew to more than $10 million ARR with eighteen people, according to a Supabase customer account. The business began with one builder and expanded as demand developed. For a founder, that’s a valuable sequence: establish a product people want, then build the team around a working business.
TypingMind went from a registered domain to a paid product in four days. Tony Dinh registered the domain on 2 March 2023 and launched on 6 March. His account of the launch reports about $22,000 in the first seven days, primarily from one-time purchases. He made existing AI models more convenient to use and quickly found customers willing to pay for that improvement.
Twenty months after launch, Dinh and his growing team had reached $1 million in revenue over the preceding twelve months, with a subscription product for teams contributing to the business. The first paid release gave him a starting point for developing both the product and its business model.
Photo AI reached substantial usage with a solo developer behind the product. Pieter Levels’ launch archive dates Photo AI to February 2023. By his August 2024 conversation with Lex Fridman, he described it generating a million or more photos a month. That’s meaningful operating scale roughly eighteen months after launch. He also explained how he used customer downloads and favourites to assess improvements to the generated images.
These founders used AI to deliver a service customers wanted: answers from documents, a better interface to models, or generated photos. Each found a focused way to turn the technology into a business. Today’s coding agents add another advantage for founders pursuing that approach: AI can help build the software as well as deliver the service customers buy.
The companies one or two people can build now
The strongest opportunities often begin with a narrow, recurring problem in an industry the founder already understands.
An agency owner might build a client reporting service. Customers connect their accounts, the software gathers the numbers, and AI drafts a report for review. An operations consultant might turn a spreadsheet process into a subscription product that collects documents, identifies missing information and follows up. A property specialist might build a portal for enquiries, maintenance requests and owner updates.
Other good candidates include proposal tools for a particular trade, specialist research subscriptions, customer support agents with carefully defined permissions, and software that turns messy supplier documents into structured records.
The business doesn’t even need to sell AI. A useful booking system, membership platform or industry-specific customer portal can use AI extensively during development while presenting customers with a familiar, simple product.
I look for three things: a customer we can reach, a problem they encounter repeatedly, and a first version small enough to build and evaluate quickly. A founder who knows the buyers can be much more valuable at this stage than another person writing code.
That is where I start helping. We work out what people will pay for, what the product must do to earn that payment, and which attractive ideas can wait.
What $15,000 a month actually gets you
Development services through Blueprint Ventures Inc. start at $15,000 USD a month. At that rate, you work directly with me across product planning, design, development, testing, deployment and launch.
You can bring a business idea without a finished technical specification. I help shape the first release, choose the architecture, design the experience, direct the AI agents and take responsibility for the implementation. The person helping clarify the vision is also the person delivering against it.
In a conventional arrangement, those responsibilities might pass between a product lead, designer, frontend developer, backend developer, infrastructure engineer and tester. For a suitable project, I can cover that work with AI assistance. You get one person who understands how the whole product fits together, with far less time spent briefing people and reconciling their decisions.
I’ve been the programmer, designer, CTO and CEO. I founded my first company at fifteen, grew it to more than 500 paying subscribers and sold it. Later, I led fifteen engineers at Datadog. That experience across software and business is what lets me judge which work can be accelerated, which assumptions need testing and where a seemingly simple request has consequences elsewhere.
For a narrowly scoped customer portal, a first month might include agreeing the product and pricing structure, establishing the visual design, setting up the application and release pipeline, and getting the first complete customer journey into a test environment. A customer could sign up, create an account, complete the main task and try the payment flow. We’d use the remaining time to address what we learn from reviewing it together.
We agree your first month’s scope before starting, around the customer journey that matters most to your product. You see working software several times a week and can make decisions while changes are still inexpensive.
The retainer includes my development tools and equipment. Product hosting and third-party services, including AI usage inside your product, are agreed separately. You own the code and infrastructure from the start. The FAQ sets out the costs and terms.
The practical value is having the capabilities needed to move a focused product forward without hiring each capability as a separate job.
What the separate pieces take to build
A SaaS product is a collection of connected parts. Looking at those parts makes the change in development speed easier to understand.
The ranges below are illustrative planning estimates for a modest web product with clear requirements, familiar technology and established services for authentication and billing. They cover implementation and focused review of each task, with recruitment, approval queues and external-provider lead times budgeted separately. A day means a working day, and a week means five working days.
Both columns assume access to templates and managed services. The conventional column assumes development without coding agents; the current column adds an experienced developer directing AI, inspecting the result and correcting it. The comparison shows where that extra implementation capacity can shorten the work.
| Piece of the product | Conventional approach | With AI and experienced direction today |
|---|---|---|
| Product brief and first-release scope, with customer input available | 3–10 days | 1–3 days |
| Starter logo and branding: refined mark, colours, typography and usage examples | 3–10 days | 0.5–2 days |
| Website design: responsive layouts for a handful of key pages | 5–15 days | 1–3 days |
| Frontend marketing website: 4–6 pages from approved design and copy | 5–15 days | 1–3 days |
| An additional content page using the existing design | 0.5–2 days | 1–3 hours |
| A modest feature such as saved filters or CSV export | 3–10 days | 0.5–2 days |
| CI/CD pipeline: automated checks, preview deployment and a rollback path | 2–5 days | 0.5–2 days |
| Initial database and API for a small, understood data model | 5–10 days | 1–3 days |
| Sign-up, login, account recovery and basic organisation permissions using an established provider | 3–10 days | 1–3 days |
| User onboarding: welcome flow, setup steps and initial emails | 3–7 days | 0.5–2 days |
| Standard subscription payments: checkout, billing portal, payment events and access updates | 3–10 days | 1–3 days |
| Customer dashboard: a few screens to view and manage the customer’s own records | 10–20 days | 2–5 days |
| Administrator dashboard: account search, status changes, permissions and an activity log | 5–15 days | 1–4 days |
| One documented third-party integration with ordinary error handling | 3–10 days | 1–3 days |
| One bounded AI feature, such as document summarisation, with an initial evaluation set | 5–15 days | 2–5 days |
| Transactional email, basic product analytics and error monitoring | 2–5 days | 0.5–2 days |
| Initial automated checks for core journeys and access controls | 3–10 days | 1–3 days |
| Operating notes and a basic handover guide | 2–5 days | 0.5–1 day |
The website rows separate designing the screens from implementing them. The dashboard estimates include connecting to an existing backend; they assume the underlying data model is already agreed. The branding estimate is for a useful launch identity. A full naming exercise or extensive brand research is a different scope.
I turn these estimates into a launch plan by identifying which tasks can overlap, sequencing the dependencies and allowing time to test the complete customer journey. Complex billing, legacy migrations, enterprise access requirements and unfamiliar integrations get their own estimates. That gives us a schedule built around your product’s actual work.
The biggest improvement is often the next change. Once the design, database and delivery process are established, adding a page or extending a familiar workflow becomes a small piece of directed work. A request you make in the morning may be ready to try that afternoon.
Why the whole process moves faster
AI compresses the work between decisions. An agent can inspect the existing application, implement a change across several files, run checks, identify a failure and work through a correction. That makes it useful for a complete development task, beyond generating a fragment of code.
It also reduces the cost of exploring alternatives. We can try two onboarding flows before committing to one. I can put an actual dashboard in front of you while we’re still discussing what customers need to see. A design decision becomes easier when you can use the result.
Earlier advances gave small companies important building blocks: cloud hosting, payment services, open-source frameworks and reusable components. Agents help carry out the work of assembling and adapting those blocks. The person directing them can stay focused on the product while moving between design, implementation and verification.
For the kinds of focused SaaS products I build, my assessment is that the distance from a clear idea to something useful has never been shorter. We can put working software into customer conversations earlier, learn what people value and use that feedback to improve the next release.
How I direct AI agents through a build
I begin with the customer’s first useful outcome. For a reporting product, that might be connecting an account and receiving a report worth sending to a client. We define that journey before writing a long feature list.
Then I establish the foundations: how customers and organisations are represented, who can access which records, what a subscription unlocks and how the application will be deployed. Those decisions give the agents a consistent system to work within.
I divide implementation into bounded tasks with explicit acceptance criteria. An instruction to add billing is too vague. An instruction to activate the correct account after a verified payment event, handle repeat events safely and test cancellation gives the agent a result we can check.
Where work is independent, agents can handle separate tasks in parallel. One can implement a screen while another investigates an integration or prepares tests against agreed requirements. I size the parallel work so I can review each result, keep dependencies in order and bring the changes together into a coherent release.
I also give them ways to verify their work: automated checks, representative data, a running application and clear descriptions of the expected behaviour. Anthropic’s guidance for agentic development similarly emphasises planning and giving agents concrete ways to check results. With those checks running throughout the build, I can review the implementation alongside evidence of how it behaves.
Finally, I test the complete customer journey. Can somebody sign up, pay, do the thing they came for and understand the result? Can a second customer access any of their records? If an external service fails, can the user recover without losing work?
We release in small increments, watch what happens and use that evidence to choose the next change. The process is described in more detail on How I work.
A working first version in weeks, with a realistic launch plan
As an illustrative planning scenario, I’d allow three to six months for a conventional custom SaaS project covering discovery, design, implementation and testing. That gives us a useful baseline for considering what faster implementation changes.
Today, I would plan a tightly scoped prototype in days and an integrated first version for feedback in weeks. A very small product with an established approach may be ready for paying users within that period. The milestone that matters is a working customer journey, with the supporting systems needed to operate it.
My published target remains 90 days for a focused MVP, with the scope and realistic timing agreed before we start. Faster implementation gives us more room within that period to test with customers, improve the experience and address the problems that appear when real people use it.
My usual total MVP budget guidance is $50,000–$100,000 USD or more, depending on scope. Within that budget, three months of development at the starting retainer costs $45,000 USD, with product hosting and third-party services agreed separately. We plan those costs together so you can see how the budget supports the launch.
You get useful software earlier, and more chances to improve it before making a bigger investment.
Building a business that stays small
Keeping the team small also means designing the product to need less manual attention. Customers should be able to sign up, get started, change their subscription and solve common problems themselves. Administrators need enough visibility to understand a failed job or account problem without investigating the database each time.
AI can help classify support requests, draft answers and prepare work for review. Routine, predictable tasks often suit ordinary automation. I choose the approach that is easiest to operate reliably, and make sure uncertain or consequential cases reach a person.
For an AI product, I also look at the cost of delivering the result. A subscription needs to cover model usage, hosting, support and the other costs of serving that customer. Usage limits, visibility into expensive requests and sensible pricing belong in the product plan from the beginning.
Managed infrastructure, self-service customer tools and specialist help where needed let a small team concentrate on the product. Starting with one or two people gives you the chance to make future hiring decisions with paying customers and a working business behind them.
That is the opportunity I help founders pursue through Blueprint Ventures. You bring the market knowledge and ambition. I help shape a product worth building, direct the AI-assisted development and take responsibility for getting it into customers’ hands.
If you have a software business in mind, tell me about your project. A few paragraphs about the customer, the problem and your budget are enough. I’ll help you work out what the first version should be and what it will take to build it.