Agentic Systems Fellowship · Cohort I

Build AI for
the agentic web.

A live programme for working software engineers. Learn how agents work, build systems that ship, and solve company tasks with people building the agentic web.

Live learningEvenings & weekendsFor working engineers

Company-task & hiring partners

500+ AccioJob hiring partners

Explore the wider network
§01  The window

The web is being rebuilt
for agents.

For thirty years the web waited for a click. That contract is being rewritten, right now, by the largest companies in commerce, payments and AI.

The agents are BEING SHIPPED
agent

ChatGPT agent and Operator: models that think and act. Research, bookings, browser work — done with you watching.

agent

Muse: A personal agent that shops from the posts you've liked, books flights, clears your inbox. Commerce from context.

agent

Instinct: a personal assistant you text or call, connected to your email and messages. Raised $1B at a $10B valuation in September, from Sequoia, Benchmark and Coatue.

The rails are BEING LAID
rails

Agentic Commerce Suite and the Agent Toolkit: agents that pass payment credentials securely, merchant registration for machine payments, MCP plugins into billing.

rails

Intelligent Commerce: tokenized credentials issued to agents, with spending controls and authentication at the network layer.

rails

Universal Commerce Protocol and agentic storefronts on ChatGPT, Gemini and Copilot. Your catalog and checkout, callable by an agent.

The fault lines ARE EMERGING
fault

Amazon blocked Meta's Muse from shopping. Agents without shared identity, permissions and commerce protocols don't get to transact.

No one can stop an idea whose time has come. The next decade of web engineering is building agent-to-merchant-to-agent trust. New problems demand new tools.

The next interface isn’t another screen.
It’s an agent that gets things done.

Identity & trustThe open webAgent protocolsCommerce rails

Agents that search, verify, bid, buy, deliver and create on our behalf need engineers who can hold the whole system at once — and the safety in between. That is what this fellowship is for.

A pencil-sketch study of a cluttered wooden desk: stacked books, an open map, a small armillary sphere, drawing compasses and a vintage desktop computer.

§02 · Company tasks

A company’s problem.
Your implementation.

First, complete the learning phase. Then, in the final two weeks, build an individual project around a real company task—with a buddy from that company.

Fini

Customer support · Fini

Resolve the request.
Don’t just answer it.

Build a support agent that finds the right policy, checks account information, and completes a permitted action. Escalate uncertain cases with the evidence a human needs.

Grounded retrieval · Tool use · Human handoff

AIONOS

Enterprise workflows · AIONOS

Turn a service inbox
into an operational workflow.

Turn messy service emails into structured requests. Extract the intent, create and route the ticket, and leave an audit trail—with approval before consequential updates.

Structured outputs · Workflow automation · Audit trails

DrDroid

AI for operations · DrDroid

Take an alert
to an evidence-backed diagnosis.

Investigate a service incident using logs, metrics and recent deployments. Explain the likely cause, cite your evidence, and propose a runbook action for a human to approve.

Tool orchestration · Context · Explainable investigations

AutoLLM

AI security · AutoLLM

Find the failure.
Then strengthen the guardrail.

Red-team an AI application for prompt injection and data leakage. Build adversarial tests, add protections, and measure both the attacks you stop and the legitimate requests you still allow.

Red teaming · Guardrails · Evaluation

1 buddyfrom the company
5–10 studentseach building independently

Shared context. Individual work.

Your buddy gives the group its task overview and is your company contact during the final two weeks. You build your own solution—not a team submission—and share your version with the company on Demo Day.

  1. 01Company briefing
  2. 02Your individual build
  3. 03Buddy feedback
  4. 04Demo Day

§03 · The programme

Understand the model.
Build the system.

Before your company project, learn to build with coding agents, understand the models underneath, and turn them into reliable software. These four blocks form the learning phase.

01 · Engineer with AI

Build software with coding agents.

Write the specification, guide the agent, and review what it changes. Work in real repositories without handing over engineering judgment.

  • Specs, rules files, code review & tests
  • Model choice, cost & when not to delegate
  • Python, FastAPI & streaming AI features
  • Claude Code
  • Codex
  • Python
  • FastAPI
  • GitHub

You’ll buildA tested feature in an existing repository, plus an AI-powered service.

02 · Understand the model

Train a small GPT from scratch.

Make the model less mysterious. Build the core pieces, watch it learn, and understand how training choices change its behaviour.

  • Tokens, embeddings & attention
  • Pre-training, fine-tuning & preference tuning
  • Sampling, loss curves & hallucination
  • PyTorch
  • Hugging Face

You’ll buildA tiny GPT you can train, sample from and explain.

03 · Build the agent

Make agents reliable enough to ship.

Build the loop yourself, then learn what frameworks buy you. Add the context, controls and measurements that make an agent useful in practice.

  • Tool use, workflows & LangGraph
  • Context, memory & RAG with citations
  • Permissions, sandboxing, security & human approvals
  • Evaluations, deployment, tracing & spend limits
  • LangGraph
  • PostgreSQL
  • OpenAI
  • Anthropic

You’ll buildAn agent and a retrieval system, with an evaluation suite and observable behaviour.

04 · See the bigger picture

Understand the agentic web.

See how agents connect to tools, services and one another—and where trust, identity and permission become engineering problems.

  • MCP, agent-to-agent communication & browsers
  • Identity, delegated access & commerce protocols
  • Data protection & the limits of autonomous action
  • MCP
  • Stripe

You’ll practiseConnecting agents to the wider ecosystem, and explaining the trade-offs.

Approximately ten weeks

Learn by building.

Live teaching, hands-on tasks and feedback across the four blocks. Your cohort moves forward together; task results do not block your progression.

The final two weeks

Build for a company.

Take on an individual company task with an employee buddy. Share your own implementation with the company on Demo Day.

Twelve weeks in all. Live sessions in evenings and weekends, with independent building between sessions.

§04 · The fit & selection

Working engineers.
Builders by habit.

At least one year of engineering experience. You already write software; you want to learn how to build with AI.

Bring your engineering foundations.

Web development and DSA in at least one language, plus SQL and databases. Python and the server layer are covered in the programme.

Be ready to write code, review it, take feedback and build independently.

The admissions process.

  1. 01
    Apply

    Tell us about your engineering experience.

  2. 02
    Interview with the AccioJob bench

    DSA, problem-solving and CS fundamentals. No AI experience required.

  3. 03
    Both sides decide

    Join a cohort of thirty working engineers.

§05 · Career opportunities

Four companies to build with.
A much wider network to meet.

Fini, AIONOS, DrDroid and AutoLLM are both company-task and hiring partners. They see your individual work on Demo Day.

Your job-opportunity pool also includes the full AccioJob partner-company network. It doesn’t stop at the four companies you build with.

500+AccioJob partner companies

Job opportunities across the wider network, alongside the fellowship’s four hiring partners.

Meet the company-task partners

Across AccioJob’s hiring network

§06 · Questions

The practical things.

Who it’s for, how it works,
and what you’ll leave with.

What is the Agentic Systems Fellowship?

The Agentic Systems Fellowship is a 12-week live agentic AI programme by AccioJob for working software engineers. You learn to build software with coding agents, train a small GPT from scratch and make AI agents reliable enough to ship, then build an individual project on a real task from Fini, AIONOS, DrDroid or AutoLLM.

Who is the fellowship for?

Working software engineers with at least one year of engineering experience. You should know web development and DSA in at least one language, plus SQL and databases. Python and the server layer are covered in the programme.

Do I need AI experience?

No. Bring your software engineering foundations. Understanding models and building with agents are what you come here to learn.

Can I do this with a full-time job?

Yes. The live programme runs in evenings and weekends, with independent building between sessions.

How are the twelve weeks organised?

Approximately ten weeks cover the four curriculum blocks through teaching and hands-on tasks. The final two weeks are for your individual company project and Demo Day.

What will I learn and build?

Four learning blocks: building software with coding agents, training a small GPT from scratch, making agents reliable enough to ship, and understanding the agentic web. Along the way you build a tested feature in an existing repository, an AI-powered service, a tiny GPT you can explain, and an agent and retrieval system with an evaluation suite.

Which tools and technologies are covered?

Claude Code, Codex, Python, FastAPI, GitHub, PyTorch, Hugging Face, LangGraph, PostgreSQL, OpenAI, Anthropic, MCP and Stripe.

Is the company project a team assignment?

No. One company buddy works with a group of 5–10 students, but each student builds independently and shares their own implementation with the company on Demo Day.

What if a task doesn’t go well?

You get feedback and continue with your cohort. Task results do not hold up your progression through the programme.

How does admission work?

Apply with your engineering experience, then interview with the AccioJob bench on DSA, problem-solving and CS fundamentals; no AI experience is required. If both sides decide to go ahead, you join a cohort of thirty working engineers.

Which companies are hiring partners?

Fini, AIONOS, DrDroid and AutoLLM are both company-task and hiring partners, and see your individual work on Demo Day. Your job-opportunity pool also includes AccioJob’s network of 500+ hiring partners.

§07 · Ask your AI

Get a second opinion.

Pick a question and open it in the assistant you already use. It reads this page and answers for your situation.

Your question

I'm a software engineer considering the Agentic Systems Fellowship by AccioJob: https://acciojob.com/agentic-systems-fellowship. Read the page, then ask me about my experience and goals before telling me whether it's a good fit and what I should know before applying.

Cohort I · Thirty seats · Begins 28 November

Don’t just watch
the agentic web happen.
Build it.

Learn the foundations. Build the systems. Put your own work in front of companies.