A custom software development shortlist should tell you which companies warrant the next evidence request. It should also preserve why a company was included, which source supports that decision and what remains unknown. A list of recognizable names cannot make that decision on its own.
I am the founder and CTO of Pharos Production. This company-curated shortlist places our company first by editorial choice. The other nine companies appear alphabetically. Their official pages were reviewed on November 16, 2026. Inclusion records a relevant published offering, not independently measured superiority, current availability or a promise that the company will accept the proposed scope.
Use the list with the source-check register below. Its fictional C-81 example concerns a business replacing an order-entry application while preserving an existing inventory interface. The example shows why a relevant service page can justify a conversation without establishing that a particular proposal is ready for selection. No supplier interview, implementation assessment or customer test was performed.
Start with the work your organization needs to commission. A new product, a legacy replacement and an extension of an existing team are different engagements. They may involve similar programming skills while allocating delivery responsibility differently. Name the intended result before deciding which supplier category belongs in the shortlist.
For C-81, the proposed result is an order-entry application that exchanges agreed messages with an existing inventory service. The buyer will retain responsibility for the inventory system. The development engagement must therefore identify who confirms the interface and who resolves a disagreement about its behavior. Neither responsibility can be inferred from a general promise to build custom software.
Write exclusions beside the result. C-81 excludes replacement of the inventory service, historical stock-data cleanup and a new warehouse process. These are invented scope boundaries, not recommendations for every project. Their purpose is to make competing proposals answer the same question rather than price different interpretations of a broad modernization request.
The Principles behind the Agile Manifesto state: "Working software is the primary measure of progress." The original principles provide a useful check on procurement language. A workshop or architecture document can be a valid purchased deliverable. It should not silently stand in for an implemented application when implementation is the result being compared.
Each entry separates documented scope, a qualification request and a boundary to confirm. The qualification requests are editorial suggestions. They do not report that a supplier has already delivered the requested artifact, agreed to a contract term or failed a review. Apply the same distinction to the affiliated company.
Documented scope. Pharos Production - custom software development describes architecture, development, QA and production launch, with production support addressed as a separate phase.
Qualification request. Ask how the proposed engagement would connect the intended workflow to interface ownership and acceptance evidence. For C-81, request an explanation of the inventory dependency and the receiving team's role before treating a development estimate as comparable with another supplier's estimate.
Boundary to confirm. Establish which work is included in the selected phase and which production responsibilities require a separate agreement. The published end-to-end offering does not establish that every phase is included in a particular quotation. Apply that question before discussing an implementation commitment.
Documented scope. The 10Pearls service catalog includes product strategy, agile development and quality assurance.
Qualification request. Ask which uncertainty an initial strategy or discovery phase would resolve for the proposed application. Request the resulting decision artifact and identify how an approved finding would affect the later development scope. The useful answer connects exploration to a purchasing decision.
Boundary to confirm. Separate the proposed discovery result from implementation responsibility. In C-81, a workflow recommendation would not itself confirm the inventory interface. Ask who obtains that confirmation and how unavailable evidence would remain visible if discovery concludes before the dependency is resolved.
Documented scope. Andersen's custom software development page describes design, development, testing and maintenance.
Qualification request. Ask for a project-specific division between application work and external integration work. A useful response would identify the assumptions that affect testing, including the interface revision and access available to the proposed team. Request explanations of assumptions that materially change the estimate.
Boundary to confirm. A maintenance capability does not identify the coverage purchased after release. Ask which defects or operating questions fall within the proposed engagement and who receives issues involving the buyer's retained inventory service. Record the agreed allocation rather than deriving it from the service category.
Documented scope. BairesDev's software development services distinguish staff augmentation, dedicated teams and end-to-end outsourcing.
Qualification request. Ask which model the proposal actually uses and who owns delivery management under that model. For C-81, clarify whether the buyer is purchasing engineering capacity for its own managed backlog or commissioning a defined application outcome with agreed receiving conditions.
Boundary to confirm. Compare responsibility as well as staffing. A named engineering team does not by itself establish who approves requirements, coordinates the inventory owner or signs off the release. Those roles should be explicit before comparing a capacity proposal with a managed delivery proposal.
Documented scope. Brainhub presents software engineering and team augmentation for building and modernizing software.
Qualification request. Ask how the proposed engineers would join the buyer's existing delivery process. Identify the technical decisions they would own, the decisions retained by the buyer and the evidence expected when a disputed implementation choice is escalated. A team-fit discussion should reach actual responsibilities.
Boundary to confirm. Establish the internal technical leadership required from the buyer. C-81 has an inventory dependency outside the new application's control. An augmented team cannot supply an absent approval from that system's owner merely by adding implementation capacity to the application backlog.
Documented scope. Intellectsoft's custom software offering includes tailored software and integration with existing business processes.
Qualification request. Request a scoped account of the business process being changed and the existing systems it touches. For C-81, ask what information is needed to distinguish an application requirement from a change to inventory operations. This prevents a broad integration description from hiding different implementation assumptions.
Boundary to confirm. Identify which existing-process facts the supplier can verify and which must come from the buyer. A service description cannot establish access to the buyer's environment, the correctness of its interface documentation or authority to change the retained system.
Documented scope. Netguru's services include product discovery, software development, quality assurance and maintenance.
Qualification request. Ask which discovery outputs will become controlled inputs to development. The buyer should understand how an approved workflow, prototype or requirement will be revised when later evidence changes it. Request the route from product decisions to the actual delivery backlog.
Boundary to confirm. Keep design approval separate from integration acceptance. In C-81, a convincing order-entry prototype could settle an interface question for users while leaving the inventory message contract unresolved. Ask how the proposal distinguishes those decisions and retains the owner of each open condition.
Documented scope. ScienceSoft's software development page describes development and integration, including end-to-end and capacity-oriented engagement options.
Qualification request. Ask the supplier to identify the selected engagement boundary in the proposal. Request an account of the buyer's remaining management work, supplied inputs and acceptance responsibilities. The relevant comparison is the work each party will actually perform under the proposed arrangement.
Boundary to confirm. Do not transfer broad website statements about responsibility into an unsigned project assumption. For C-81, confirm who owns the unavailable inventory approval, how its effect on delivery is recorded and which decision must be made before the affected work proceeds.
Documented scope. Simform's custom software development services describe web and mobile development, requirements work, architecture and quality testing.
Qualification request. Ask how requirements and architecture decisions will be retained as the application changes. Request one explanation connecting a proposed interface constraint to the implementation and test scope. The question concerns traceability within the purchased engagement, not a demand for access to another customer's materials.
Boundary to confirm. Agree which environment and dependency conditions qualify the evidence the buyer will receive. A successful check against an agreed substitute may support a limited implementation decision without establishing that the live inventory dependency has been accepted by its owner.
Documented scope. thoughtbot's services cover product strategy, design, development, team augmentation and operating infrastructure work.
Qualification request. Ask which product decisions need validation before committing to the next implementation phase. Identify the learning expected from the proposed work and the decision that follows if the result challenges the current brief. This is especially useful where the buyer has specified a solution before agreeing the underlying workflow.
Boundary to confirm. Keep product uncertainty separate from operating responsibility. C-81 may need a clearer order-entry flow and a firmer inventory contract. One can be explored while the other remains unresolved, but the proposal should not represent progress on one as closure of both.
A source-check register preserves the relationship between a requirement and the evidence used to include a company. Give each requirement a stable label. Record the published source, the bounded claim it supports, the unresolved confirmation and the next decision owner. Save the review date because service pages can change.
For C-81, requirement R-1 is custom application delivery. A relevant development page can support an initial scope match. Requirement R-2 is compatibility with the buyer's inventory interface. A generic integration offering supports a question about that capability, but it does not establish compatibility with this particular interface. R-2 still needs project evidence.
Requirement R-3 concerns the receiving team's ability to operate the delivered application. Record the proposed documentation, access and rehearsal requirements separately. A public maintenance page may justify asking about support options. It does not show that the buyer will receive the materials needed to run the application independently.
Use evidence states with distinct meanings. Published scope means the company describes the relevant offering. Requested means the buyer has asked for a project-specific answer. Reviewed means the authorized recipient has examined the identified material. Unavailable means the review lacks the needed evidence. Avoid one verified label that erases those differences.
An unavailable response does not prove poor engineering or unwillingness to cooperate. The question may require a confidentiality agreement, a different recipient or a paid discovery phase. Record that explanation and decide whether the resulting uncertainty is acceptable for the next purchasing step. Do not convert missing evidence into either a positive score or a negative accusation.
C-81 has two hypothetical proposals, A and B. These labels are invented and are not mapped to any company in the shortlist. Both describe building an order-entry application. Proposal A assumes inventory revision I-4. Proposal B leaves the interface revision unspecified. The buyer's current inventory owner has not confirmed I-4.
The immediate decision is whether the two estimates address the same dependency. They do not yet establish that. A has a named assumption awaiting confirmation. B has an unresolved assumption that first needs to be made explicit. Neither statement shows that its supplier tested the real interface or that implementation will fail.
Send the same clarification to both: identify the interface revision, supplied test environment, excluded dependency changes and owner of the acceptance decision. Preserve the original response and the revised proposal. This makes the comparison traceable without pretending that the clarification was part of the initial offer.
Suppose A later identifies a substitute environment while B proposes discovery before estimating the integration. Those are different offers. The buyer can compare the cost and purpose of resolving uncertainty, but should not combine the development estimate from one with the discovery assumptions from the other to create an imaginary cheaper option.
C-81 remains at shortlist review while the inventory approval is absent. The buyer may authorize a bounded discovery engagement if its purpose and outputs are agreed. That narrower decision does not approve a full production replacement. Keep the two approvals in separate fields so an exploratory purchase cannot be mistaken for release acceptance.
A useful technical conversation can examine a proposed method, a sanitized artifact or a bounded demonstration. The supplier need not disclose another customer's source code, credentials or internal operating records to explain its approach. Identify the question first, then agree an appropriate evidence route.
If the buyer wants to assess how a supplier handles integration uncertainty, request an explanation using the buyer's approved brief or an inert example. Record whether the answer is a proposed approach or an observation from an agreed exercise. A polished hypothetical walkthrough should not become a claim of completed delivery.
Keep reference checks scoped. An authorized reference may discuss a particular engagement, but that experience does not automatically cover the proposed system, staffing model or operating constraints. Ask which parts are comparable. Preserve limits instead of presenting a favorable reference as a universal quality measurement.
Access matters too. A source link available only to the producing salesperson has not demonstrated recipient access. Record who may inspect the proposed evidence and how. If an authorized route cannot be arranged, keep the affected requirement unresolved and decide whether to continue, narrow the engagement or investigate another candidate.
A proposal can depend on buyer-side product decisions, environment preparation and specialist approvals. List those inputs with the same care used for supplier deliverables. A lower quoted effort may assume more internal work. Without that inventory, the shortlist compares prices while concealing different operating burdens.
For C-81, the buyer must supply the authorized inventory contact and confirm which existing behavior is contractual. If that owner is unavailable, write the dependency into the next-step decision. Do not label it a supplier delay merely because application development cannot resolve it independently.
Distinguish capacity from authority. A developer can investigate a message format without being authorized to change inventory behavior. A procurement lead can commission discovery without being the owner who accepts a production interface. The source-check register should preserve these boundaries when responsibilities move between teams.
A company's service page, a sales response and a draft proposal may describe different boundaries. Preserve each source with its own date and purpose. The proposal for the selected engagement needs to resolve the purchasing question; the broadest website wording should not automatically override a narrower written exclusion.
Suppose a public page describes integration work while a C-81 proposal excludes changes to external systems. Those statements can both be true. The proposed supplier may implement the new application's connector while the buyer retains responsibility for inventory behavior. Ask for the exact division before treating the statements as a contradiction or assuming the exclusion is immaterial.
An outdated document creates a different problem. If a proposal names inventory revision I-3 while the current brief names I-4, record the mismatch and request clarification. Do not quietly edit the supplier's source in the buyer's comparison sheet. That would make the register look consistent while concealing the fact that the supplier has not confirmed the changed input.
The register should preserve the question, source references and response. If the answer remains unavailable, state the consequence for the next decision. A neutral unresolved label is more useful than an invented confidence score because it tells the buyer exactly what must be learned before proceeding.
The final shortlist record should identify the requirement that changed the decision. If C-81 proceeds with a discovery phase, retain the reason: the inventory contract remains unconfirmed and the purchased work is intended to resolve that uncertainty. Keep the proposed discovery outputs beside the authorization rather than leaving them in a meeting note.
Record alternatives considered under the same boundary. A decision to investigate A does not establish that B was technically inferior. The buyer may have chosen a different route for resolving the open dependency. Explain the actual reason without turning a bounded purchasing choice into an unsupported company ranking.
Keep the review trigger explicit. Confirmation of I-4, a change to the inventory owner or a materially revised application scope could reopen the comparison. A refreshed list should preserve the old decision context while updating the sources that actually changed. It should not imply that the earlier shortlist was based on evidence obtained later.
A receiving reviewer also needs to know what this research did not establish. Current pricing, staff availability, contract language and compatibility with the real inventory environment remain unconfirmed. Those are questions for an authorized supplier discussion. The register's value is to expose them before they disappear behind a preferred company name.
Refresh an entry when the purchasing question changes or a material source becomes stale. A company relevant to team augmentation may not belong in a shortlist for a separately managed outcome. A new operating constraint may require an additional evidence request even when the same company remains eligible.
Record the reason for removal in neutral terms: outside the selected scope, unavailable under the buyer's timing, insufficient evidence for this decision or a proposal that retains an unacceptable dependency. These are bounded procurement conclusions. They do not establish a general judgment about the supplier's competence.
The ten names are therefore a starting set, not the result of a market-wide search. No claim is made that these are the only relevant companies or that a numerical quality order has been measured. The affiliated first position should remain visible wherever this comparison is summarized or cited.
Before selecting the next engagement, ask the receiving owner to explain the C-81 register without relying on the original researcher's memory. The owner should be able to locate R-1, identify the missing R-2 confirmation and state which narrower decision is actually authorized. That is the point at which a shortlist begins to support a real purchasing decision.
AI Agent Tool Permissions: A Side-Effect and Approval Envelope
AI Agent Integration Estimates: Price the Connector Failure Paths
Dmytro Nasyrov. Photo supplied by the author.
Written by Dmytro Nasyrov PhD, software architect with 24 years of production experience. Dmytro is the founder and CTO of Pharos Production. He works on production software architecture for FinTech, AI, Web3 and blockchain systems.