Skip to main content

What's new at Coodesh: Q3 2026 (July to September)

In the third quarter, Coodesh started working inside the AI assistants your team already uses, added a new way to assess developers who code with AI, and became more accessible and fairer for the people taking the assessment.

Connect your AI assistant to Coodesh

Claude, ChatGPT, Copilot, and Codex can now connect to your Coodesh account through a simple authorization flow, with no API key (MCP connector). From the assistant, your team can look up jobs, applications, assessments, results, careers, and credits, and also handle day-to-day tasks: moving candidates between stages, creating jobs, assessments, and questions, and generating invitation links.

What this solves for you:

  • Recruiters work where they already are, without switching tools to answer a question or set up a process.

  • Questions like "who passed the cutoff score on this assessment?" get answered in seconds, with no spreadsheet export.

  • Control for your IT team: each person sees and revokes their own connections, admins can cut any connection that reaches the company, authorization expires after 90 days, and offboarded users lose access immediately. Every action taken by the assistant is logged under the name of the person who connected it.

Tech Assessment: remote IDE questions to assess developers who code with AI

Candidates code in a full development environment, right in the browser, with nothing to install. You decide per question whether AI use is allowed, and you can adjust that policy for each assessment.

What this solves for you:

  • You assess developers the way they actually work today, with AI assistants, instead of a test that pretends those tools don't exist.

  • Reviewers get the final code, the commit history and, when allowed, the conversation with the AI agent. You see the reasoning, not just the result.

  • Project and IDE questions accept the template as a .zip file, so you don't have to publish the test on GitHub.

Integrity: faster reviews and rules that fit your process

Integrity review is now clearer and adapts to how strict each process needs to be:

  • Guided review: the session recording player now has chapters per question and findings marked on the timeline. The integrity card counts each event type and takes you straight to it, and coding and SQL answers show how the code was written.

  • More accurate signals: new signals include the test window losing focus and a large block of text inserted without pasting. Faces in the background, such as a coworker, a picture, or the TV, no longer count as a second person.

  • Configurable rules: each assessment sets the weight of each signal, starting from ready-made profiles or your own thresholds. Before saving, a simulation shows how many attempts would change level. Recalculation only changes the integrity level, never the score.

  • Optional blocks when the candidate leaves full screen, when the webcam fails, or when a second monitor is detected.

What this solves for you:

  • Reviewing a suspicious case takes minutes, not hours of watching recordings.

  • Fewer good candidates penalized by false alarms.

  • The right level of rigor for each process: an internship program and a critical senior role don't need the same rules. The platform flags the signals, and the decision about what they mean stays with your team.

End-to-end accessibility

Accessibility features now live in a single menu, with reading adjustments, Brazilian Sign Language (Libras) translation, and the new Face Control, which moves the cursor with head movements and clicks when the person pauses over a target. The candidate journey was reviewed with screen readers and keyboard navigation: the timer is read out in full, shortcuts jump to the prompt, the questions, and the submit button, and screens work down to 320 px wide. Assessments can also offer extra time to anyone who requests an accommodation.

What this solves for you:

  • Truly inclusive hiring processes that support affirmative roles and disability inclusion programs.

  • Candidates with disabilities take the same assessment as everyone else, without your team having to adapt it manually.

Learn more: Accessibility menu

Question appeals and targeted reopening

Candidates can challenge a question during the test and move on to the next one. Your company decides in the new appeals tab, with three options: give credit to that candidate only, void the question for everyone (with a preview of which scores change), or reopen that candidate's assessment. Reopening is also more precise: you can release only the selected questions, and the report flags the redone answer alongside the previous score.

What this solves for you:

  • An ambiguous question no longer becomes a reputation problem: you fix it transparently and know in advance which scores will change.

  • A candidate's unexpected issue is solved without retaking the whole test, with everything recorded in the report.

Skills Intelligence: careers that follow your structure

  • Positions and tracks can be moved between tracks and between careers without losing requirements, maps, or people, and a simulation shows what changes before you confirm.

  • Careers now have version history, salary ranges per country, and spreadsheet import of people.

  • In skills maps, bulk invitations accept up to 1,000 rows, endorsements work alongside self-assessments, and the skills matrix can be exported per map or consolidated.

What this solves for you:

  • Restructuring a department doesn't mean rebuilding the career: it follows your actual org chart.

  • Multi-country operations work with local salary ranges.

  • Skills data comes out ready for workforce planning.

AI Interviewer: more control over the experience

Candidates choose the interview language before hearing the first question. Your company decides what they see afterward (their own answers, the full transcript, or feedback per criterion), and reviewers see exactly what was published. If the connection drops, the interview resumes at the pending topic, without repeating what was already answered.

What this solves for you:

  • Interviews in the candidate's language across multi-country hiring.

  • Transparency with candidates, at the level your company defines.

  • A dropped connection doesn't become a lost interview. And the final decision stays with the recruiter.

Governance and security for large teams

  • Activity log: an audit trail with 129 tracked actions, including assessments, invitations, results, users, plan, credits, SSO, and integration keys. It has filters, time windows, and CSV export.

  • An account organized your company's way: custom profile fields, your own level scales, members limited to the jobs they work on, and access requests from people with an email from a verified domain, approved by an admin. The "Access profile" option shows the platform as a regular user sees it.

  • Verified CNPJ (companies in Brazil): verification with the Brazilian tax authority runs automatically, and approved companies get a "Verified" badge.

What this solves for you:

  • Audit-ready for compliance and data protection: you know who did what, and when.

  • Adoption within your company's security policies, with no exceptions for IT.

  • Large operations scale without the admin becoming a bottleneck.

Other improvements

  • From assessment to job: a published assessment becomes a job posting in three steps, with data prefilled and AI-suggested copy, and applicants land straight in the assessment. For jobs with external applications, a new tab shows clicks and interest.

  • Pause and scheduling: your company sets how much pause time the assessment allows, and candidates can schedule a date and time, with a Google or Outlook invite and an email reminder.

  • Recordings that don't get lost: live microphone levels, chunked uploads that resume without re-recording, presentations up to 50 MB, and a framing check before the test.

  • Results: option to hide per-question scores and to schedule result release for a few days later or a fixed date.

  • Export center: choose the columns, preview, and get larger files by email.

  • The assessment list shows a funnel by status, and search finds names with or without accents.

  • Background Check: each check shows the product purchased and delivers the right report.

  • Email templates with error-proof variables and a send history.

  • User records with more than one manager, department, and academic background.

  • Datalake version 2, faster and with skills map and integrity event data for your BI.

  • Much faster screens: in the assessment editor, switching tabs dropped from 2.4 s to 130 ms, and the dashboard is 2 to 3 times more responsive.

What changes in your day-to-day

  • Section scores now count unanswered questions as zero. An abandoned test with 12 correct answers out of 60 no longer shows up as 75%. The exception is an adaptive test ended by the engine itself.

  • Assessment exports now go into the report center queue, and the "report ready" email links to the file.

  • New assessments are created with accessibility on and 25% extra time. If you prefer a different value, adjust it in the settings.

  • In skills maps, the N/A option was removed and the "Intermediate" level is now called "Practitioner".

For admins and integrators

  • Datalake version 1 will be shut down on 02/28/2027. Migrate to version 2 and store records by id (upsert). See the guide.

  • MCP connector: connections older than 90 days must be redone, and the get_answer tool is now called get_assessment_answer.

  • Assessment export via API now returns a ZIP file instead of a JSON list.

  • Permissions: listing jobs and recalculating integrity on delivered results now require dedicated permissions.

  • Privacy: user profile data is only shown to people linked to the company, the talent feed no longer includes contact details or resumes, and resume links expire after 15 minutes.

  • Spreadsheets built on exports: the skills matrix now has one row per skill, and the Completeness Index moved from a 0 to 1 scale to 0 to 100.

Questions about any of these updates? Reach out to your Coodesh point of contact or explore the articles in this Help Center.

Did this answer your question?