For in-house engineering teams
For companies whose developers already write production code and use, or plan to use, Claude Code, Codex or Cursor. We help your team agree shared standards, change how review works, build your policy into the tools' own controls and set a baseline, so you can see whether delivery is improving. Work is scoped and quoted after an assessment.
We reply within an hour during UK business hours.
.png?height=28&name=Gunnebo%20Logo%20(1).png){kind=small}
.png?height=28&name=Closer%20Still%20Media%20(1).png){kind=small}
.png?height=28&name=Ecommpay%20Logo%20(1).png){kind=small}
.webp?height=28&name=Tussell%20Logo%20(1).webp){kind=small}
-4npmo6jxod.png?height=28&name=iconsolutionsrgb%20(1)-4npmo6jxod.png){kind=small}
AI for engineering teams is our help for in-house developers to run Claude Code, Codex and Cursor well as a team, covering configuration, shared instructions, review, governance and measurement. Google's DORA research reported in 2025 that AI adoption is associated with higher delivery throughput and also with more delivery instability.
Each developer installs a tool with personal settings, and nobody can say which models, permissions or connected tools are in use.
Approved tools are set up through each vendor's admin controls, with sign-in through your identity provider and permissions agreed for the team.
Each developer prompts the agent differently, so the same repository gets different conventions depending on who asked.
Instruction files, skills and hooks are agreed per repository and kept under version control, and your engineers own what they say.
Agents produce more change than reviewers planned for, so review either slows everything down or gets waved through.
Agreed rules for how agent changes are sized and who approves them, with hosted AI review used as a second reader and the merge decision kept with your engineers.
Licence numbers and accepted suggestions are reported, though neither shows whether delivery got faster or more stable.
Delivery and stability measures are taken before the pilot, so the team compares like with like as use grows.
Whoever holds a seat turns on cloud agents, and nobody has settled where code, prompts and secrets go.
A readiness check and an agreed security position come before any agent works unattended on your code.
Six pieces of work, taken in the order your team needs them. Each applies to Claude Code, Codex and Cursor alike, and you can use more than one of them.
We help you decide whether to standardise on one tool or allow several, then configure sign-in, provisioning, policy and permissions in each vendor's own admin settings. The right choice depends on your repositories, hosting platform, plans and data terms.
We draft the instruction files, skills, subagents and hooks each repository needs, agree them with your engineers and put them under version control, with the controls that must hold kept in settings rather than in prose.
We help you set how agent changes are sized, what a person must approve before production and where hosted AI review fits, so review keeps pace without becoming a formality.
We tier work by risk and turn your written policy into managed settings, permissions, hooks and pipeline checks, using published UK government engineering standards as reference points.
We set up delivery and stability measures before the pilot and add each vendor's usage analytics alongside them, so you can judge the effect on delivery rather than on activity.
Before any agent runs unattended, we work through where code, prompts and secrets go, which repositories and networks it can reach, and who may switch it on.
We look at your tools, plans, repositories, review practice and current measures, then agree the scope and a fixed price with you.
We draft the repository standard, review rules and policy controls for your engineers and security team to amend and approve.
One team works under the new standards on real work while the baseline is measured. You decide whether to widen it based on what the measures show.
The configuration and standards extend to further teams and repositories, with training on your own work and changes made where the pilot showed gaps.
We compare delivery and stability against the baseline, revisit the controls as the vendors change their products, and agree what happens next.
Claude Code is Anthropic's and Codex is OpenAI's. Cursor says it was acquired by SpaceX in August 2026, and it lists SpaceX Super Intelligence (SpaceXSI) among the model providers it supports. The tools are listed alphabetically. Plan packaging changes often, so we check each vendor's current documentation during the assessment.
All three offer single sign-on from their team plans (Claude Team, Codex Business, Cursor Teams). SCIM provisioning is listed for Enterprise plans only on all three.
Each has a way to apply settings centrally, through an admin console, device management or managed configuration files, and each supports shared instruction files. Anthropic and Cursor both say in their documentation that instruction files guide the agent and are not a security control.
All three support hooks that run at points in the agent's work. On each, a hook that errors or times out can let the action through unless that is handled deliberately, and each vendor says its own controls are not a complete security boundary.
Audit records are listed for Enterprise plans only, and each vendor limits what they cover. Usage analytics and OpenTelemetry export are available, with different plan requirements on each tool.
We agree this with you during the assessment. Your engineers keep the decisions about your code, and we do the configuration, drafting and training around them.
Decide which tools are approved, own the content of the instruction files and make every merge decision.
Sets the data and access rules, decides who may switch on cloud agents and which repositories they may reach.
Configures the admin controls, drafts the standards and policy for your approval, trains your engineers on their own work and sets up the baseline.
Ready to Build is for people without a developer background who build with AI. Safe to Ship reviews one AI-built app. Our Engineering Team subscription is our engineers doing the work for you.
Most teams start with a conversation about their tools and what they want to change. If the people building are not developers, a different service will suit you better.
Tell us about your team, your tools and what you want to change. We scope the work after an assessment and give you a fixed price before anything starts.
Our diagnostics help growing organisations decide where to invest next. The AI and Data Readiness assessment is the nearest fit while you work out what your team needs.
Ready to Build gives people without a developer background version control, automatic security checks, proper hosting and training.
Our work is fixed price, scoped and agreed before we start, and quoted in pounds.
SpotDev is an OpenAI Select Partner, is Certified in the Claude Partner Network, and is a registered SpaceX Super Intelligence Transformation Partner.
Every SpotDev developer holds Claude Certified Developer: Foundations. Some of our engineers also hold OpenAI Codex Deployment Practitioner and Codex Solutions Practitioner.
An in-house, fully remote engineering team, certified to Cyber Essentials Plus. We use AI coding tools, including Claude Code, Codex and Cursor, in our own engineering.
It is our service for companies whose developers already write production code and use, or plan to use, Claude Code, Codex or Cursor. We help the team agree how the tools are configured, what shared instructions and review look like, which controls enforce your policy, and how you will measure the effect on delivery. We do not build your software or place our engineers in your backlog.
It depends on where your repositories are hosted (each vendor's hosted code review supports different platforms), which plans and admin controls you need, your existing contracts and data terms, and whether you want cloud agents. Some teams use more than one. We set out the trade-offs for your team and you make the call.
Each vendor documents admin controls for sign-in, provisioning, policy, permitted extensions, spend and usage analytics, and the plan each control needs differs by tool. Anthropic, OpenAI and Cursor each say in their documentation that these controls are not a complete security boundary on their own, so we pair them with branch protection, least-privilege access and human review.
Not on its own. Anthropic describes CLAUDE.md files as context rather than enforced configuration, and Cursor says AI guidance should not be your only security control. Codex reads AGENTS.md files before it starts work. These files make the agent more consistent. Anything that must hold goes into managed settings, permissions, hooks or pipeline checks, and hooks need care too, because on all three tools a hook that errors can let the action through unless that is handled deliberately.
It can be, once a few things are settled. Cursor says Cloud Agents are the only Cursor feature that requires it to store code. Anthropic says organisations with Zero Data Retention enabled can't use Claude Code's cloud sessions. OpenAI switches Codex Cloud off by default for Enterprise workspaces. We work through where code, prompts and secrets go, which repositories and networks an agent can reach and who may switch it on, before any agent runs unattended.
Take a baseline of delivery and stability measures before the pilot, because seat counts and accepted suggestions do not show delivery. In the UK government's trial of GitHub Copilot and Gemini Code Assist (November 2024 to February 2025), participants reported saving 56 minutes a working day, while only 39% reported committing code the assistant suggested. METR's early-2025 trial, which found experienced open-source developers took longer with AI tools on their own mature repositories, is now marked out of date by its authors, who say the current effect cannot yet be measured reliably. Measuring your own team is the sound approach.
Usually. A study of 806 open-source repositories that adopted Cursor found activity rose for about two months while static analysis warnings and code complexity stayed higher, and its authors identify quality assurance as the main bottleneck. DORA suggests larger batches of code are harder to review. We help you set how agent changes are sized, what a person must approve before production and where hosted AI review fits as a second reader.
Start from published examples. The Home Office's engineering standard requires AI-assisted outputs to be reviewed and approved by a human before reaching production, and says teams retain full accountability for AI-assisted code. The Ministry of Justice has published an engineering AI governance framework for other organisations to adapt. We help you tier work by risk and turn the policy into controls in the tools you use.
Ready to Build sets up AI coding tools for a builder with no developer background. Safe to Ship is a review of one AI-built app by our engineers, signed off by our CTO. This service is for teams of professional developers deciding how to run the tools across their work. If you need extra engineering capacity, that is our Engineering Team subscription.
That depends on the size of your team, the number of tools and repositories, and how far you want to go. We scope the work after an assessment and give you a fixed price, agreed before we start and quoted in pounds. Request a Quote to begin.
Related:
SpotDev is an OpenAI Select Partner, is Certified in the Claude Partner Network, and is a registered SpaceX Super Intelligence Transformation Partner. Certification in the Claude Partner Network does not imply endorsement by Anthropic, and partner status does not imply endorsement by OpenAI or SpaceXSI. Claude Code, Codex and Cursor are trademarks of their respective owners.
