Ten Questions for Any Vendor Using AI Coding Tools

What to have in writing before a software vendor lets AI tools or agents near your code. Tick what is true, read your score, copy or print the list.

Before a software vendor lets AI coding tools or agents near your code, get ten things in writing: who owns the output, whether your code trains a model, who holds which credentials, who reviews and logs each change, what an agent may do without asking, and who pays when work is redone. StepTo is a 15-20 engineer nearshore team in Belgrade (CET). StepTo publishes four relevant commitments. An NDA and an IP assignment are signed before any engineer starts, and all IP goes to you. Engineers work in your repository and your tools from day one, and you own the code at all times. Every change is reviewed by a second engineer before it is merged. Engineers work on company-managed laptops with disk encryption and two-factor sign-in. It publishes nothing on its own use of AI tools, so run us through the list as you would any vendor.

The ten questions

0 of 10

Many open questions. Get the missing answers in writing before AI tools touch your code.

Tick a row only when a good answer is in writing: in the proposal, the contract or the vendor's own documentation. Nothing you tick is stored or sent anywhere.

Ownership and IP

  • A good answer
    The contract assigns all work product to you, states that this includes output produced with AI tools, and is signed before work starts.
    A red flag
    Ownership is described only on the vendor's website, or the vendor reserves rights in tools, prompts or generated components.
  • A good answer
    A written list of approved tools and model providers, with a rule that adding one needs your approval.
    A red flag
    "We use the best tools for the job" with no list, or a list that the vendor may change without telling you.

Data and model training

  • A good answer
    A written statement for the vendor and for each listed tool that your code and prompts are not used for training, with the plan tier or terms it applies to.
    A red flag
    "It depends on the tool" with no table, or an answer that covers the vendor but not the tools it resells or wraps.
  • A good answer
    Named providers, the regions they process in, a retention period for each, and a deletion route when the engagement ends.
    A red flag
    No retention answer, or a promise that data is "secure" with no provider, region or period named.

Access and least privilege

  • A good answer
    Personal accounts, no shared logins, scopes limited to the repositories and tasks in the engagement, and a named owner who removes access and shows you it is gone.
    A red flag
    A shared service account, a token with organisation-wide scope, or no named owner for offboarding.

Review and audit trail

  • A good answer
    Branch protection that requires an approving review from a second person, applied to every change whatever its origin, visible in your own repository.
    A red flag
    Review is described as "AI-assisted" with no named human approver, or reviewers are allowed to approve their own changes.
  • A good answer
    A convention such as a commit trailer or pull request label, plus tool and agent activity logs you can request, kept for a stated period.
    A red flag
    No record at all, or a record that sits only in a vendor tool you cannot read.

Agent and MCP permissions

  • A good answer
    A written default of read-only access, an allowlist of approved MCP servers and tools, and an approval step before any new connection or wider scope.
    A red flag
    Agents run with the engineer's full permissions, or engineers may install any server or plug-in they like.
  • A good answer
    Agents run in a sandbox without your production secrets, and any high-impact action needs a named person's approval, logged.
    A red flag
    "Fully autonomous" is offered as a feature, or agents can reach production credentials, deployment pipelines or customer data.

Billing of rework and tool usage

  • A good answer
    A stated rule that defects found in agreed acceptance windows are fixed at the vendor's cost, and a statement of whether tool or token charges are inside the rate or passed through, with a cap.
    A red flag
    Rework billed as new hours, or usage charges that appear on the invoice without being in the proposal.

Run StepTo through the list

Bring the rows you could not tick to a 15-minute intro call, or start from the terms we publish.

What do the published standards say about agent permissions?

Two sources behind rows 5, 8 and 9, both read on 2026-11-20. The OWASP Gen AI Security Project page LLM06:2025 Excessive Agency names the root cause as one or more of "excessive functionality", "excessive permissions" and "excessive autonomy". Among its preventions it says to "Utilise human-in-the-loop control to require a human to approve high-impact actions before they are taken."

The Model Context Protocol page Security Best Practices recommends a "progressive, least-privilege scope model" and lists an "expanded blast radius" among the risks of a token that carries broad scopes. For local MCP servers it recommends running them in a sandboxed environment with minimal default privileges. A vendor's answers to rows 5, 8 and 9 should look like these controls applied to your engagement.

How do larger vendors publish their answers?

Globant and Andela publish written answers to rows 1 and 3, and Andela also to row 8. SoftServe and HTEC publish certification lists instead. We link them as examples, not as endorsements, and a published answer is not a verified one.

  • Globant reports in the AI Pods FAQ that, for AI Pods for Software, you own what gets built and that "Your data stays your exclusive property and never trains anyone's model", as of 2026-11-20.

    Source: globant.com/ai-pods

  • Andela reports five governance gates (Approve, Isolate, Control access, Review and monitor, Expand deliberately), "No model training" on client data and that you own the code, as of 2026-11-20.

    Source: andela.com/deploy/managed-ai-solutions

  • SoftServe lists "AI Security and Governance" as one of 19 service lines, and lists ISO 27001 and SOC 2 Type 2 among its certifications, as of 2026-11-20.

    Source: softserveinc.com/en-us/services, softserveinc.com/en-us/about-us

  • HTEC lists ISO 27001 certification and SOC 2 Type II attestation reports on its quality page, as of 2026-11-20.

    Source: htec.com/quality-first

On certifications, SoftServe and HTEC win: both publish ISO 27001 and SOC 2 Type 2 or Type II. StepTo makes no certification claim on this page. If your procurement requires a certificate, ask every vendor on your list, StepTo included, to produce it.

How should you use the checklist?

Send the ten rows with your request for proposal and ask the vendor to answer each one in the proposal itself. An answer that exists only on a call is a question still open. The score is a prompt for what to ask next, not a verdict on a vendor.

For the commercial and contract terms around these rows, run the same vendor through the twelve-check vendor scorecard. StepTo's own answers to common procurement questions are on the due diligence answers page, and the practices behind them are on the engineering practices page. To turn your answers into a brief, use the software RFP template. If the work itself is an AI project, read what a forward deployed engineer does. For StepTo's terms, see the pricing page and the staff augmentation page.

For background on the risks behind the rows, see our posts on agentic engineering and outsourcing, AI agents with too much access, shadow AI and LLM security and usage-based billing for AI coding tools.

Frequently asked questions

What commitments does StepTo publish that apply to AI-assisted work?

An NDA and an IP assignment are signed before any engineer starts, and all IP goes to you. Engineers work in your repository and your tools from day one, and you own the code at all times. Every change is reviewed by a second engineer before it is merged. Engineers work on company-managed laptops with disk encryption and two-factor sign-in. StepTo does not publish a list of AI tools or a model-training policy, so ask for both in writing.

Who owns the output when a vendor's engineers use AI tools on your code?

Under StepTo's published terms, you own the code. An NDA and an IP assignment are signed before any engineer starts, and all IP goes to you. Engineers work in your repository and your tools from day one, and you own the code at all times.

Where do StepTo engineers work, and on what devices?

Engineers work in your repository and your tools from day one, and you own the code at all times. Engineers work on company-managed laptops with disk encryption and two-factor sign-in.

Can I run StepTo through this checklist before I sign?

Yes. Bring the rows you could not tick and ask for the answers in writing. It starts with a 15-minute intro call. If there is a fit, a 60-minute discovery call follows. StepTo's published answers are on the due diligence page.

Get in touch

Or reach us directly

Office hours

Monday – Friday09:00 – 18:00 CET
WeekendClosed

Sent on a weekend: we reply on the next business day.

Contracting party: StepTo d.o.o. Company details are on the Imprint page.

Send us a message

StepTo's Belgrade team replies within one business day.

It starts with a 15-minute intro call. If there is a fit, a 60-minute discovery call follows. Engineers can typically start within one to two weeks of first contact; a full team usually takes two to four weeks.

* required

Your details are handled as described in our Privacy Policy.

Performance-led engineering

Want senior engineers who move work forward, not just tickets?

Work with accountable, English-fluent professionals who communicate clearly, protect quality, and deliver with a steady operating rhythm. Cost efficiency matters, but performance is why clients stay with us.

Delivery signals · senior engineering team
Senior ownership
Lead-level
Delivery rhythm
Weekly
Timezone overlap
CET
1 teamaccountable for outcomes, communication, and execution