FAQ
The questions people ask before signing anything.
Including the ones that are awkward to answer. If something you need to know isn't here, ask us directly and we'll add it.
Shark Labs works on fixed-price projects quoted after a short paid discovery, with clients owning all code and infrastructure from the first commit. Projects are billed in milestones, retainers monthly with thirty days' notice and no minimum term. The studio is founder-led, based in Bengaluru, and works remotely with clients worldwide.
Money
How much does a project cost?
AI features start at $3,000, marketing sites at $1,500, MVPs at $6,000 and retainers at $900 a month. Every range is published on the pricing page along with what moves the number — the biggest factor on AI work is the state of your data, not the complexity of the model.
Is the price fixed or an estimate?
Fixed, after discovery. The ranges we publish are indicative; the number in your written scope is specific and doesn't move unless you change what's being built. If we misjudged the effort, that's ours to absorb — which is exactly why we insist on paid discovery before quoting.
Why is discovery paid?
Because it's real work: reading your codebase, testing your data, and writing an implementation plan. Free discovery is either superficial or it's priced into other clients' projects. It's $400–1,800 depending on scope, it's credited against the project if you proceed, and you keep the plan either way — including if you take it to another studio.
What are the payment terms?
Projects are 40% to start, 30% at the agreed midpoint and 30% on delivery, with longer projects split into more milestones so neither side is ever far ahead. Retainers are billed monthly in advance. Indian clients are invoiced in INR with GST as applicable; overseas clients in their own currency.
What if we run out of budget mid-project?
Tell us early. Because we build in slices with something working from week one, there's almost always a defensible stopping point where you have a usable thing rather than a half-finished one. That's a deliberate property of building in that order, and it's much easier to use if you raise it in week four than in week eleven.
How we work
Who actually writes the code?
The founder, Yash Mittal. Shark Labs is deliberately small — there's no account layer, and the person on your scoping call is the person committing. Specialists come in for design and for platform work that genuinely needs them. More about who that is.
How long does a project take?
An AI feature into an existing product: three to six weeks. A RAG system: four to eight. A marketing site: three to six. An MVP: eight to sixteen. All measured from the end of discovery, and the variance is nearly always scope rather than pace.
How often will we hear from you?
A working demo every week, and a written update alongside it. No status meetings for their own sake — if a week's update is 'this took longer than expected and here's why', that's the update, sent on time.
Can you work with our existing team?
Yes, and it's often the best shape: we build the first version and your engineers pair on it, so the knowledge stays after we leave. What we're not is staff augmentation — if you need someone inside your sprint taking tickets from your PM, hiring is the right way to buy that.
What timezone do you work in?
India Standard Time. That overlaps comfortably with Europe, the Middle East and Asia-Pacific, and gives a few hours with the US east coast. US west coast clients get a handover rhythm instead of a shared afternoon — some teams find that faster, some find it frustrating, and it's worth being honest about which you are.
Ownership and risk
Do we own the code?
Entirely, from the first commit. It lives in your repository, deploys to your cloud accounts, and uses your API keys. There's no licence, no hosting arrangement you have to keep paying for, and no component that only runs on our infrastructure. Leaving means telling us, not migrating.
What happens if we want to stop working with you?
On a project, you pay for the milestones completed and take everything built so far. On a retainer, thirty days' notice with no penalty. Because everything already runs on your accounts, there is no handover project to negotiate — which is the point of setting it up that way.
How do we avoid becoming dependent on you?
Insist on the handover being a deliverable rather than a courtesy: documentation written for the engineer who inherits it, an eval suite your team can run without us, everything on your own accounts, and a walkthrough with your engineers. All four are standard here. If a studio resists any of them, that tells you whether the dependency is accidental.
What if the project goes wrong?
Fixed price means an underestimate on our side is our cost, not a change request. If the thing we build doesn't meet the written scope, we fix it. What no contract can cover is the risk that the feature ships correctly and users don't want it — which is why we push hard for the smallest version that answers that question first.
Do you sign NDAs?
Yes, routinely, before discovery if you'd prefer. We'll also sign your MSA rather than insisting on ours, provided it doesn't contain unlimited liability or an indefinite non-compete — both of which are unreasonable to ask of a small studio and we'll say so rather than sign and hope.
Is our data used to train AI models?
Not on the enterprise API tiers we build on — Anthropic and OpenAI both contractually exclude API traffic from training by default. Where that isn't sufficient, because of a regulator or a customer contract, we build on open-weight models running on infrastructure you control. That costs more per query and we'll show you the difference before you decide.
Whether to hire us at all
When should we hire in-house instead?
Once software development is permanently on your roadmap rather than a project — practically, past about eighty engineering hours a month, or two ongoing workstreams. Above that you're paying studio rates for what is effectively an employee. The full comparison, with salary numbers, is here.
How do we vet a development agency?
Four questions. Ask to see a system they built that's live in production, and ask what its accuracy is — if there's no number, there's no eval suite. Ask what happens when the model provider deprecates a version. Ask who owns the infrastructure. And ask them to describe a project they turned down; a studio that has never declined work is telling you how it scopes.
What kind of work do you turn down?
Staff augmentation, pure visual redesigns of sites that work fine, anything requiring fabricated credibility, and projects where the real problem isn't the one we'd be hired to solve — a surprising share of AI enquiries turn out to be data or process problems. We'd rather say that on the first call than three weeks in.
You're new. Why take the risk?
It's a fair question and the honest answer is a smaller first project. Take a $600–1,800 discovery, or a single well-scoped feature, and judge the work rather than the pitch. That's a real evaluation with a bounded downside, and it's how most engagements here start regardless.
Do you have client testimonials?
Not published ones yet, and we'd rather say that than commission a wall of quotes. What exists instead is detailed case studies of real client work, each with the architecture described and a live URL, and the founder's prior public work — an npm package with sixty-plus releases and several live applications. Testimonials go up when there are real ones to publish.
Next step
Still got a question?
Ask it directly. If it's a good one we'll add it to this page, which is where most of these came from.