Agentic Associate
- LocationCape Town, Western Cape, South Africa
- Work typeHybrid
- EmploymentFull-time
- Posted16 November 2026
- SourceBsky's careers page (BambooHR)
Last checked36 min ago: still listed on Bsky's careers page. First seen 36 min ago.
At a glance
- Experience1+ year
- Skills and toolsAWS, Salesforce, Python, Java, TypeScript, Git
Read from the ad by RoleFinder, with rules and an AI reading; the description is the source.
About the role
Agentic Associate , BlueSky Digital Solutions
The role
The Agentic Associate builds the pieces that let Claude agents work safely inside a client's systems, so you'll write the tool descriptions an agent reads, the tests that prove each tool does only what it says, and the checks that catch an agent doing something it shouldn't.
BlueSky is an Anthropic partner, a Salesforce Summit Partner and an AWS Advanced Consulting Partner, and we deliver AI, AWS and Salesforce work for large South African enterprises, from banks to telecoms and retail, where we're getting backend services ready for agents to use, so in your first year you'll be doing real agent work inside a large enterprise's own systems, which is rare this early in anyone's career.
Reports to: The architect or senior engineer on your workstream
Location: Hybrid from Cape Town or Johannesburg
Level: Associate
What you'll do
- Write the tool manifests agents read. For each backend service we open up to an agent, you describe what the tool does, what it may touch, and which identifiers it takes from the case rather than from the request, so an agent can't be talked into changing the wrong record.
- Write the contract tests. Prove each tool does exactly what its manifest says and nothing more, including the retry that must never pay twice and the payment that can only be released against the reference it belongs to.
- Build into our toolkit. Our skills and agents ship as shared plugins the whole team installs, and once a few of your fixes have gone through review you'll write smaller additions yourself on a branch, with the tests that prove they work. When one gets something wrong on a job you fix it for the whole team with a pull request that carries its tests, rather than patching your own copy.
- Prepare work for senior review. Package each piece so a senior can approve it quickly: what was asked, what you and Claude built, what you checked, and what still needs a decision.
- Test the guardrails around Claude Code. Write the hook test files that prove a dangerous command gets stopped and keep the sandbox and automatic security review settings the client's own engineers rely on in good shape.
- Run the poisoned-document test. Hand an agent documents with hidden instructions in them and check that nothing moves, because untrusted input, sensitive data and the power to act should never meet without a person confirming first.
- Run the agentic assessments. Our Claude practice reads a client's Claude setup, from seats and identity to Claude Code settings, cost, security and how their agents are built, without changing anything, and you'll run those reads and prepare the findings for the senior who signs them off.
What we're looking for
Must have
- One to three years in a technical role, such as software, data, cloud or platform consulting, or something equivalent
- You've built something real with Claude Code or the Claude API in the last few months, and you can show us and tell us what you checked before you trusted it
- Working knowledge of AWS or Salesforce, enough to find your way around a client's environment, and you don't need both
- Clear written English: you can explain what you did and why in a few lines
- Comfort reading and writing code in at least one language, Python, Java or TypeScript, and with Git
Nice to have
- Claude Certified Developer Foundations, or the plan to sit it in your first six months with our support
- Experience with MCP servers, evals or agent loops
- An interest in how agents get misused: prompt injection, data leaking out, an agent acting on something it shouldn't
- Exposure to banking or another regulated industry
Who you are
The tools change every few weeks, so we care more about how you think than about any one framework. We're looking for someone who's properly curious about how agents behave and why they misbehave, who enjoys being the one who spots what's wrong, asks the obvious question out loud when nobody else will, and still says so when the plain option is the right one.
That's our four values in practice, Embrace Difficult, Make It Better, Pay It Forward and In It Together, and the last one is why every piece of your work gets a second pair of eyes before it reaches a client, which is how we look after each other and the client, never a judgement on you.
Where this goes
In the first six months, success means you've shipped tool manifests and tests that went through the client's own review, you can take a piece of work from brief to review without someone checking each step, and you've passed Claude Certified Developer Foundations.
Growth path: Strong associates move toward designing agent guardrails themselves, and then into the Forward Deployed Agentic Engineer role, running a workstream end to end.
What you get
- Leave: 20 days annual leave
- Flexibility: hybrid from Cape Town or Johannesburg, three days a week either at our office or at the client and the rest remote
- Learning: your own Claude seat from day one, Claude Certified
- Developer Foundations paid for by us, an AWS or Salesforce certification if you want one, time to build on real client work, technical training, and a mentor who'll guide you, challenge you and back you
- Before you start: depending on the client, expect background checks
How we hire
A short intro call, then a session where you walk us through something you've built with an agent and what you'd change, and we aim to give you an answer within two weeks of that session. Claude helps us sort applications by role, and a person reads and decides on everyone. We keep your details for 12 months only to run this search, so tell us if you'd rather we didn't.
If you'd like to see how we work before you apply, have a look at our culture page at https://bsky.co.za/culture/. To apply, go through the posting on our careers page, which asks three short questions: something you've built with an AI agent recently, what you do when an agent hands you work you think is wrong, and whether you'd rather build, run or lead.
Get open roles ranked against your CV
Add your CV once. RoleFinder ranks the open roles where you want to work, plus remote ones, against it, shows your best 12 with a one-line reason each, and emails you new matches once a week. The file is read once by AI and not kept; employers never see it.
Sign in with Google or email to start. How we handle your CV