Most guides to software development staffing models rank the same suppliers on one axis: who has the best developers, at what rate. That's the wrong first question. A product isn't built by developers alone, and the model that covers all three disciplines is rarely the one that lets you walk away cheaply. You're choosing the shape of a team, not a vendor. Pick by the mistake you can least afford.
Key takeaways
Software development staffing models are the different ways you can assemble the people who build a product: in-house hiring, open freelance marketplaces, vetted marketplaces, dedicated-team agencies, and subscription/on-demand teams. Each differs in how many disciplines it covers and how easily you can reverse the decision.
A shipped product needs three jobs done: someone to write it (dev), someone to design how it works (UX), and someone to prove it actually works (QA). Almost every staffing model on the market sells you one of the three well and treats the other two as your problem.
So judge models on two axes, not on rate cards. First, coverage: how many of the three disciplines does this model actually put on your team? Second, reversibility: what does it cost you, in time and money, to change your mind six weeks in? Those two decide more than any hourly figure, and neither shows up in a comparison table of rates.
Design is the discipline that goes missing first. A 2020 Nielsen Norman Group survey found the most typical researcher-to-designer-to-developer ratio was 1:5:50, and only half of respondents had even one designer for every ten developers — the other half had fewer [R24]. Before you've chosen a single vendor, design is already the starved seat at the table. Most models will happily keep it that way.
In-house hires. You own the capability forever, which is the point when the work is your core advantage. It's also the slowest and most expensive to start and the hardest to unwind: every seat is a full recruiting loop and a payroll commitment.
Open freelance marketplace (Upwork-style). The cheapest headline number. You buy access to people and nothing else: vetting, management, integration, and quality control all stay on your desk. Fine for a bounded task, punishing at team scale.
Vetted marketplace (Toptal, Arc, Lemon). You're buying vetting, billed hourly, one role at a time. Toptal organizes talent into seven categories and matches specialists to you [R1][R2]; QA exists only as a sub-page under Developers [R3], and Arc lists no QA role at all [R8]. You still manage delivery yourself. Rates sit in a separate post. See our Toptal-alternatives cost comparison for the numbers, and our vetted-vs-open marketplace breakdown for which kind fits the task.
Dedicated-team / outsourcing agency (Netguru, BairesDev, Acropolium). A managed, multi-role team that a vendor runs for you. This is the model that genuinely bundles disciplines: Netguru's standard team includes a QA engineer and a product designer alongside the developers [R11]. The trade is minimum commitments, a ramp period, and notice periods measured in months [R18].
Subscription / on-demand (DevOD). A flat monthly fee per engineer, a sequential task queue with a task-by-task approval gate, cancel any time. You trade parallel horsepower for reversibility and coverage in one subscription.
| Model | Disciplines (Dev/UX/QA) | Time to first useful work | Minimum & lock-in | Who manages delivery | Cost to change your mind |
|---|---|---|---|---|---|
| In-house | All three, if you hire all three | ~75-day median technical fill [R14] | Permanent payroll | You | Restart a 75-day fill + full loop [R14] |
| Open marketplace | One at a time, you assemble | Days, if you vet fast | None | You (vetting + integration) | Low, but you own every gap |
| Vetted marketplace | Dev; UX separate role; QA thin/absent [R1][R3][R8] | ~48h to match QA [R3] | Two-week trial, short notice [R4][R6] | You | Cancel per role; you still manage [R6] |
| Dedicated agency | Dev + UX + QA in one team [R11] | 4–6 wks to sprint velocity; 6–10 to full [R11] | Minimum hours; ~2-month notice [R18] | Vendor | Notice period + 4–6 wk re-ramp [R11][R18] |
| Subscription (DevOD) | Dev + UX + QA, one subscription | First task within 5 business days | None; cancel any time | Vendor (per task) | Cancel any month; swap engineer |
Only two do reliably. A dedicated-team agency bundles Dev, UX, and QA in one managed team [R11], and a subscription model like DevOD carries all three on one subscription. Vetted marketplaces sell design as a separate role and treat QA as thin or absent [R1][R3][R8].
Here's the thesis. Vetted marketplaces do sell design, but as a separate role, separately contracted, at its own hourly rate. Toptal lists Designers as one of seven catalogue categories [R1][R2], and QA lives only as a sub-page under Developers [R1][R3]. Arc lists no QA or testing role anywhere on its pricing page [R8]. Andela has repositioned around AI engineering: its site sells AI service lines and engineer archetypes, with no UX and no QA in sight [R9]. Its own /skills/quality-assurance URL now serves AI-archetype content that never mentions quality assurance as a contracted discipline [R10]. The role didn't move. It disappeared.
The uncomfortable twist for anyone selling subscriptions: the model that most reliably bundles all three disciplines is the dedicated-team agency. Netguru's standard team ships with a QA engineer and a product designer built in [R11]. Most subscription dev vendors don't. Acropolium's seven-person subscription team includes a QA engineer but no designer [R20]; Poket Dev's subscription is a Python developer plus cloud/DevOps, with no design and no QA discipline at all [R20]. So "on-demand" usually means dev, sometimes dev plus QA, almost never design.
DevOD's distinction is being a subscription that carries all three: Dev, QA, and UX engineers on one subscription. If you want the honest cost of each discipline on its own, our UX-designer cost post and QA-engineer cost post break those down.
Ashby's platform data — 54 million applications across 93,000 jobs — puts the median time to first fill at 75 days for technical roles versus 60 for business roles [R14]. Every alternative model is, at bottom, selling you those 75 days back. Skip the search; start next week.
But matching speed and productive speed are different things. A new engineer takes an average of 33 days to reach their tenth merged pull request (as of April 2026), a figure that has dropped more than 50% since early 2024 [R15]. DX measures this across its own dataset, not split by staffing channel, but there's little reason a ramp curve would bend much depending on where the engineer came from. DX, which publishes that number, cautions that it measures ramp speed, not the quality of the work, the rework it generates, or how deeply the person understands your system [R15]. Agencies quote 4–6 weeks to sprint velocity, but the same page also says full productivity takes closer to 6–10 weeks [R11] — treat neither as settled.
You're buying a head start on getting started, not on being useful. Since the useful part takes about the same time whoever you pick, the variable worth optimizing is how fast you find out it isn't working. That argues for models you can cancel this month, not ones you're locked into while the ramp plays out.
The hourly rate is the least informative number on the page, because it prices the person and ignores the coordination. Netguru — an agency, arguing against its own retainer — tells clients to add 25–35% to the quoted figure for their own PM time, onboarding dilution, and tooling [R13]. That's overhead you pay whether or not the vendor's rate is competitive.
And it's not a vendor defect you can shop around. Microsoft studied 484 of its own developers and found they spend about as much of the week on communication and meetings (~12%) as on writing code (~11%) [R16]. That's inside one company, with no vendor boundary at all. Coordination is physics. Every organizational seam you add — a marketplace contractor, an offshore team, a timezone gap — loads more onto the line item that's already your biggest. (One offshore-vendor guide estimates 15–20% management overhead plus a similar productivity hit across large timezone gaps [R17]; that's a vendor's estimate, not measured data.) The models that need the least managing quietly win a column nobody puts on the comparison table.
Reversibility is the axis you don't feel until you're trapped on the wrong side of it.
A vetted marketplace gives you a two-week trial and a short cancellation window, but you're contracting per role and still managing the work yourself [R6]. A dedicated agency asks for minimum monthly hours and a notice period of roughly two calendar months on one vendor's terms [R18], plus a 4–6 week re-ramp if you switch teams [R11]. Andela, according to third-party reviewers (Tecla, an Andela competitor, not Andela itself), carries a 12-month minimum and a $50,000 buyout to convert an engineer early [R25]. In-house feels cheapest to exit because there's no contract, but a mishire restarts that 75-day median fill [R14] and re-runs the entire loop from the job post down. If you want the hire-vs-marketplace-vs-subscription math laid out, our subscription-engineering breakdown works it through.
A subscription is the outlier here: cancel any month, and swap an engineer under a replacement guarantee if the match is wrong. Reversibility is a feature you only learn to value after a lock-in has cost you a quarter.
In-house wins when the work is your core moat and has to stay in the building. Justin Leader, CEO of Human Renaissance, puts the risk plainly: "This is the logic that shifts every week because customers keep teaching you something. Outsource it and you are paying a vendor to learn your business well enough to be acquired instead of you." [R-LEADER]. If that's the work, hire.
Open marketplace wins for one well-defined, low-risk task you can specify tightly and vet yourself. Reach for a vetted marketplace when you need elite specialist vetting fast and will pay hourly for it. A dedicated agency is the answer for a large, multi-month, fixed-scope build where you want a vendor to own delivery end to end and you genuinely need all three disciplines in one managed team [R11][R12] — that's the coverage argument, and on coverage the agency beats us.
Now turn it on ourselves. A subscription is the wrong answer when you need deep, long-term domain ownership — hire. It's wrong when you have a single, well-scoped one-off — use a marketplace. And it's wrong when you need many people working in parallel right now, because the model is deliberately sequential: one task at a time per engineer, with an approval gate between tasks. Even "unlimited" subscription plans tend to enforce this: a design-subscription analysis (from Tapflare, itself a competing subscription vendor) notes such services usually cap you at one or two active tasks at a time [R27]. Said plainly: an agency beats us on all-in-one-team coverage, and a direct hire beats us on continuity. If either of those is your binding constraint, we're not your answer.
Collapse the whole decision to one move. Name the discipline blocking you most right now — the design you can't ship, the code you can't write, the release you can't trust. Hand that discipline one real task. Give it one cycle. Judge the output before you commit to any shape. (Here’s how to run that one-task trial without betting the roadmap.) You're not trialing a vendor; you're trialing the model, which is the thing you actually can't renegotiate later. If you want the wider concept, this sits inside the on-demand product team model.
That's exactly what DevOD's intake is built for: a 15-minute fit call and a one-task Proof of Quality before you subscribe, first task shipped within 5 business days, and you own all the IP. If the model fits, a single stream runs $3,495/mo per engineer; two engineers in parallel is $6,795/mo on the Dual stream.
Stop ranking rates. Compare coverage and reversibility, because those are the two things you can't change your mind about once you've signed.
What's the difference between staff augmentation and a dedicated team? Staff augmentation — a vetted marketplace — gives you vetted specialists billed hourly, one role at a time, while you still manage delivery yourself [R1][R6]. A dedicated-team agency runs a managed, multi-role team for you, bundling Dev, UX, and QA in one team, in exchange for minimum commitments and notice periods measured in months [R11][R18].
Which staffing model covers design and QA, not just development? Two do reliably. A dedicated-team agency bundles Dev, UX, and QA in one managed team [R11], and a subscription like DevOD carries all three on one subscription. Vetted marketplaces sell design as a separate contracted role and treat QA as thin or absent [R1][R3][R8].
How fast can each model start delivering? An open marketplace can start in days if you vet fast; a vetted marketplace matches in about 48 hours [R3]; a dedicated agency reaches sprint velocity in 4–6 weeks [R11]; an in-house hire follows a ~75-day median technical fill [R14]; and a subscription like DevOD ships its first task within 5 business days.
When is in-house still the right choice? Hire in-house when the work is your core moat and has to stay in the building — logic that shifts weekly as customers keep teaching you something [R-LEADER]. Owning the capability forever is the point when the work is your core advantage, even though it's the slowest and most expensive model to start.
Is a subscription cheaper than hiring? A subscription trades parallel horsepower for reversibility and coverage rather than being simply cheaper. A single DevOD stream runs $3,495/mo per engineer and a Dual stream $6,795/mo with no lock-in, while an in-house mishire restarts a 75-day median fill and the full recruiting loop [R14].
