AI coding tools write tidy code fast. Some of it is wrong, some of it calls things that do not exist, and the tests it writes can pass whatever the code does. This self-paced practitioner course teaches working developers to give the tool the right context, plan in small checkable steps, supervise an agent, and review and test what comes back. You practise on Tallybook, a fictional expenses app, in browser labs. About 17 hours over 6 weeks.
Bodhih AI-Assisted Software Development: Practitioner is a self-paced online course for working developers and tech leads who use AI coding assistants and agents. In about 17 hours it teaches eight skills: predicting where a tool will fail, giving it context, planning spec-first, supervising an agent, reviewing AI-written code, judging tests, keeping secrets and licence risk out, and setting team standards. Learners practise in browser labs on a fictional codebase and complete a reviewed capstone from their own work.
Facts on this page last checked by Siddharthan VGJ, Director, Bodhih Training.
Generated code is tidy by default, so tidiness tells a reviewer nothing. The usual problems sit at an edge case, in a call that does not exist in your version of a library, or in a test that would pass if the function were deleted.
An agent given a loose task and wide permissions will finish it somehow: a raised limit, a new dependency, a reformatted file. Without a scope, checkpoints and a way to stop it, you find out in the diff, or later.
Speed is easy to feel and hard to measure. Field experiments show real gains on average and wide variation between teams, and time saved while writing can come back as time lost in review and rework.
You accept suggestions every day. You will learn to forecast where the tool will be wrong before you ask, give it the context it lacks, and review what it returns against five specific targets.
You approve the pull requests and answer for the outcomes. You will learn to set review depth by risk, write a one-page team agreement and report the effect on delivery with its limits.
More of what reaches you was written by a tool. You will learn to tell a test that proves something from one that cannot fail, and to use tests to pin behaviour before a change.
You are moving from suggestions to agents that edit files and run commands. Picture your next agent run with a written scope, the least permission it needs and a checkpoint before anything hard to undo.
This is not a first programming course, and it does not teach one product’s menus and shortcuts; if you want a tour of a specific tool, the vendor’s own documentation and courses will serve you better.
Four dated findings, each linked to its source and stated with its limits.
In a Stanford University user study, participants with an AI assistant wrote insecure solutions more often than those without on four of five programming tasks, and were more likely to believe their code was secure.
Perry, Srivastava, Kumar and Boneh, “Do Users Write More Insecure Code with AI Assistants?”, ACM CCS 2023, 2023Across 80 coding tasks and more than 100 models, the models chose an insecure way to write the code 45% of the time. This is a vendor study.
Veracode press release, 2025 GenAI Code Security Report, 30 July 2025Across three randomised field experiments with 4,867 developers, those given an AI coding assistant completed about 26% more tasks. The authors note that each experiment is noisy, and less experienced developers gained more.
Cui, Demirer, Jaffe, Musolff, Peng and Salz, “The Effects of Generative AI on High-Skilled Work”, Management Science, 2025In 2021, New York University researchers gave an early AI coding assistant 89 security-relevant scenarios chosen to be risky; of the 1,689 programs produced, about 40% were found vulnerable.
Pearce, Ahmad, Tan, Dolan-Gavitt and Karri, “Asleep at the Keyboard?”, IEEE Symposium on Security and Privacy 2022, August 2021Imagine opening your next pull request knowing which lines the tool was likely to get wrong, and having already checked them. The course measures you at the start and the end on parallel forms, and asks for a capstone: one real change taken from spec to merge, with your findings and tests, scored by a reviewer before you receive the credential.
Explain how assistants and agents produce output, name six ways they fail, and sort a task into likely strong, needs context or needs your judgement before you ask.
Write a project instructions file as rules that can be checked, supply the right files for one task, and keep secrets and personal data out.
Break a feature into small steps with acceptance checks and an out-of-scope list before any code is generated, and hold the plan when the request grows.
Scope a task, set the least permission it needs, place checkpoints where a mistake would be costly, and stop or redirect a run that drifts.
Search a change for five things: wrong behaviour, made-up calls or packages, security holes, tests that prove nothing and edits outside the task.
Pin behaviour before a change, spot an expected value that was quietly edited, and ask for tests that cover the boundaries.
Decide what to send, clean first or never send, check a suggested dependency and its licence, and catch common insecure patterns in generated code.
Use AI on legacy code safely, agree team standards for AI-assisted pull requests, and report an honest number with its limits.
The pathway runs in order: a baseline assessment, eight modules (how the tools work; context; spec first; driving an agent; reviewing AI-written code; tests as guardrails; secure and compliant use; legacy code and team practice), each with a hands-on lab and boosters two and seven days later, a reviewed capstone on one AI-assisted change from your own work, and a final assessment that unlocks once the capstone is accepted.
Measured before you start and again on the same instrument at the end, so the change is recorded rather than claimed.
The baseline assessment, Form A: 20 scenario questions on all eight competencies, about 25 minutes. It is not pass/fail; it sets your baseline, and you take the parallel form at the end.
What an AI coding tool can see, where it guesses and six ways it fails, with a one-minute forecast before every request. Inspector labs: The rate limit answer and The agent's report.
Three layers of context, a project instructions file that states rules the AI can follow, and what never belongs in context. Context Lab: build the Tallybook instructions file and run three tasks against it.
A written spec before any code: small steps, acceptance checks and what is out of scope. Scenario Sim: The feature that grew, then three rounds of spec writing.
Scope, permissions and checkpoints for a coding agent, and when to stop it. Flight Deck mission: The refactor run, a coding agent in the Tallybook repository.
Five review targets for AI-written changes, including invented dependencies and design drift. Code Review Lab: The CSV export and The password reset, eleven problems to find.
Tests that pin behaviour before a change, and how to tell when an AI-written test proves nothing. Code Review Lab: The rounding refactor and The loyalty discount.
Secrets, personal data, dependencies and licences when you code with AI: what stays out of a prompt, and what to verify before you accept the output. Paste Check: Before you paste, then the Code Review Lab: The receipt scanner.
AI on legacy code, shared team standards, pull requests and measuring value honestly. Scenario Sim: The team rollout, then an agreement builder for your own team.
Take one real change through the whole method with an AI coding tool your employer approves. Hand in one document of 1,200-2,000 words plus code extracts, in five parts: (1) context: the task, the codebase and your project instructions file, with your forecast of where the tool would fail; (2) the spec: small steps, acceptance checks and what was out of scope; (3) the delegation record: scope, permissions and checkpoints, and at least one point where you stopped or redirected the tool; (4) review and tests: at least three findings in the AI-written change with the evidence for each, the tests that pin the behaviour, and one AI-written test you rejected or strengthened, with the reason; (5) security and team note: your check for secrets, personal data, new dependencies and licences, one measure of the effect with its limit, and one standard you would propose to your team. Remove secrets, customer data and anything your employer treats as confidential first. If you cannot use work code, use a personal or open-source project you are allowed to change, and say so. A reviewer scores your work on five criteria (0-4 each). Met is 12 or more of 20, with every criterion at 2 or more and Review and tests at 3 or more. Feedback comes within 5 working days, and you may resubmit once.
The final assessment, loaded from Form B: 20 parallel scenario questions, about 25 minutes. Score 70% or more (3 attempts), with your capstone accepted, for the credential.
Finishing this pathway issues a credential backed by the evidence you produced along the way. It is verifiable by link and is also issued as an Open Badge 2.0, so an employer can check it without taking your word for it.
Three layers of context and a project instructions file, built and tested in the Context Lab against three tasks on the Tallybook codebase.
Small steps, acceptance checks and an out-of-scope list, practised in the Scenario Sim The feature that grew and three rounds of spec writing.
Scope, least permission and checkpoints, flown in the Flight Deck mission The refactor run, with a debrief that shows which setting caused each result.
Five review targets and a four-pass routine, practised on The CSV export and The password reset: eleven problems to find across two changes.
Tests that mirror the code, skipped tests, over-mocking and loose comparisons, found in The rounding refactor and The loyalty discount.
A Paste Check on what may leave your machine, a review of The receipt scanner, and the Scenario Sim The team rollout with an agreement builder.
What comes with the course, and what the practice is like.
Tallybook (fictional) is a small expenses app with real-looking rules: money in whole cents, a test command, a ban on unreviewed dependencies. Every lab uses it, so by Module 5 you review changes against rules you wrote in Module 2.
Each Code Review Lab change has seeded problems and some sound lines. You mark lines, name the kind of problem and give a verdict. Notice how often the evidence for a bug sits in the test file, not in the code.
Ship It Safely, The AI Agent Deployment Playbook, The AI Use Policy and Governance Kit, Automate the Boring Parts of Your Job and The AI-Fluent Professional.
Company seats with reporting on each learner’s progress, before-and-after scores and capstone status. Module 8 produces a one-page team agreement you can adopt.
Ask about team seats| What is compared | A typical AI coding course | AI-Assisted Software Development at Bodhih |
|---|---|---|
| Focus | Generating code faster: prompts, shortcuts and tool features. | Verification: forecasting failure, review, tests and security, alongside context, specs and agent supervision. |
| Practice | Watch a demo, then follow along in your own editor. | Browser labs with seeded problems to find: a Code Review Lab, a Context Lab, an agent mission, sims and Inspector labs. |
| Tools | Built around one vendor’s product. | Tool-neutral: any approved assistant or agent. No product tour. |
| Evidence | Productivity claims with no source. | University, standards-body and vendor studies named in the module where they are used, with their limits. |
| Measurement | A completion certificate. | A baseline and a final on parallel forms and a reviewed capstone; your before and after shows on the credential. |
| Honest limit | Often a live cohort with an instructor. | Self-paced with no live instructor; you do not write and run code inside the course. Larger work happens in your own editor. |
No ratings yet. Only people who have bought this pathway can rate it - every buyer gets a one-click rating link a couple of days after their purchase.
Straight answers on time, price, tools and the credential. If yours is not here, ask us and you will have a reply the same working day.
Working developers, tech leads, QA and test engineers, and platform engineers who use, or are about to use, AI coding assistants and agents. A role lens switches the examples to your kind of work. It is not a first programming course.
None in particular. The course is tool-neutral and names no product: the skills (context, specs, supervision, review, tests, secure use) apply to any assistant or agent. The labs run in the browser on fictional examples, and the at-work tasks use whichever tool your employer approves.
Not inside the course. The labs are review and decision labs: you read short JavaScript and Python changes, mark the lines that are wrong, build an instructions file, set an agent’s permissions and make calls in simulations. You write code in your own editor for the at-work tasks and the capstone.
About 17 hours: eight modules of about 90 minutes, a 25-minute assessment at the start and at the end, and a capstone of about four hours at work. Most people finish in about 6 weeks at two modules a week. Short booster questions arrive two and seven days after each module.
You take one real change from spec to merge with an AI coding tool and hand in a short document in five parts: context and forecast, the spec, a delegation record, your review findings and tests, and a security and team note. A reviewer scores five criteria and one resubmission is allowed. If you cannot use work code, use a personal or open-source project you are allowed to change.
The reading is the same; the search is different. AI-written changes tend to fail in a few typical ways, so the course teaches five targets: wrong behaviour at an edge, a made-up call or package, a security hole, a test that proves nothing, and a change outside the task. Depth of review is matched to risk, so routine changes move fast and changes to money, access or personal data get more.
It may, and the course does not promise a figure. Published field experiments show gains on average and wide variation. Module 8 teaches you to measure the effect on your own team in five steps and to report it with its limits, including rework and review time.
Yes. When your capstone is accepted and you pass the final assessment (70%, with up to three attempts), you receive the Bodhih AI-Assisted Software Development credential with an Open Badge 2.0, a verify page and a PDF that shows your before-and-after result. It is a Bodhih credential, not a university qualification, and it does not prepare you for any vendor exam.
No. The course is self-paced with no live sessions. A person reviews your capstone and writes feedback within five working days.
Yes. Until 10 January 2027 the price shown on this page is a launch price, about half the regular price. After that date the regular price applies.
Yes. Company seats are $79 each, with reporting on progress, before-and-after scores and capstone status. Write to solutions@bodhih.com or call +91 99000 11601 for seats or a rollout.
The course is $99 for an individual, or $79 per seat for companies. You can start today with the baseline assessment, then make your first one-minute forecast in Module 1. For a team, write to solutions@bodhih.com or call +91 99000 11601.
Starting records where you stand today. That baseline is what the re-measurement is compared against, and it is never rewritten afterwards.
For IT, security, risk and engineering staff: map where AI can be attacked, layer defences, limit agents and write a control plan, practised in browser labs.
Set up, supervise and test AI agents without code, in the hands-on Agent Flight Deck lab.
Lead your team’s AI use safely and lawfully: the 2026 rules, three Scenario Sims and a real AI register entry.
Measure your team first, plan each person’s growth after, or bring a trainer or coach in. Bodhih’s sister companies cover the rest of the journey.