iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
“H.I. First” is a call to keep human judgment in charge of AI-assisted work—not proof that 99% of AI workflows fail. Masum Billah’s article offers that figure as rhetorical framing, without a study or measurement to support it. Its practical argument is that people should define the problem, supply context, choose trade-offs, verify results, and remain accountable while AI handles bounded execution tasks.
What “H.I. First” means
In Masum Billah’s manifesto, “H.I. First” means human intelligence and engineering judgment should lead the work. AI can help carry out defined tasks, but people should decide what needs to be built, what constraints matter, and whether the result is fit to use.
![The Microsoft Copilot Bible: [3 in 1] A Beginner's Guide to Harness AI in Office 365 to Automate Workflows, Generate Content, Analyze Data, and 10x Your Productivity](https://m.media-amazon.com/images/I/41+5Qv4pd2L._SL160_.jpg)
Billah argues that generic AI output can reflect an “algorithmic average,” while engineers add value by applying domain rules, system constraints, optimization goals, and judgment. That is the author’s framework, not a finding established by a controlled study.
Is it true that 99% of AI workflows fail?
The manifesto does not substantiate the 99% figure with data, a study, a defined sample, or a measurement period. It should be read as headline rhetoric, not as a verified failure rate for AI workflows generally. The article also raises questions about code quality and strategic depth; those prompts do not establish that either has declined.
#1 Best Overall
That distinction matters: a forceful statistic can make an argument memorable, but without a method and evidence it cannot tell readers how often workflows fail, what counts as failure, or which kinds of AI use are involved.
Where human judgment belongs in the workflow
Define the problem and its constraints
People should set the objective and provide the context an AI system cannot be assumed to know: domain rules, existing architecture, security requirements, operational limits, and the consequences of getting a decision wrong. Better problem definition is more useful than treating prompt wording as the only lever for quality.
Choose among strategic trade-offs
AI can help explore options, but people must weigh competing goals such as performance, maintainability, cost, compatibility, and risk. The right choice depends on priorities and circumstances, not just whether a generated answer looks plausible.
Review the output and own the decision
Generated code, designs, and other artifacts need review proportionate to their impact. A person or accountable team should accept, revise, or reject the work and retain responsibility for approval and release. Delegation does not transfer that accountability to the model.
Rank #3
Delegate bounded work, not the whole outcome
A practical division of labor follows the manifesto’s distinction between execution and governance. Use AI for a task with a clear scope and reviewable result; keep people responsible for requirements, consequential choices, verification, and release decisions.
- Before delegation, specify the intended result, relevant constraints, and what a correct output must satisfy.
- Choose tasks for which a reviewer can realistically inspect and test the result.
- Check generated work against project requirements and applicable security practices before incorporating it.
- Keep a traceable record of generated artifacts, review, revisions, and approval where the project’s risk calls for it.
- Do not treat fluent or polished output as evidence that it is correct.
This is a workflow design principle, not a guarantee that review will catch every defect or that AI assistance will improve results by a predictable amount.
What NIST guidance adds for software teams
NIST’s SP 800-218A, published in July 2024, supplements the Secure Software Development Framework with practices for generative AI and dual-use foundation models. It provides a security-focused basis for considering AI-related risks in software development; it does not validate the manifesto’s 99% claim.
NIST’s National Cybersecurity Center of Excellence also describes human-directed generative AI assistance in its Notional Reference Model for DevSecOps for Demonstration of NIST SSDF. The model illustrates oversight, review gates, traceability, and accountable approval of generated artifacts. It is an institutional workflow example, not evidence that any particular process prevents all failures or delivers a measured improvement.
NIST notes in SP 800-218A that “AI model and system development is still much more of an art than an exact science, requiring developers to interact with model code, training data, and other parameters over multiple iterations.” That is a reason to treat development as iterative and controlled—not to assume an output is dependable without evaluation.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Questions to ask before adopting an AI workflow
- Who sets the requirements? A named person or team should define objectives, constraints, and acceptance criteria.
- What is being delegated? Keep the scope clear enough that the task and its result can be assessed.
- How will the output be tested and reviewed? Decide what evidence is needed before generated work is accepted.
- How are security and sensitive information handled? Consider those risks as part of secure software development, not as an afterthought.
- Can the work be traced? For consequential changes, preserve the connection between generated artifacts, review, and approval.
- Who is accountable for release? Make clear who can approve the result and remains responsible for the outcome.
The right controls depend on the task and its risks. The central point of the H.I. First argument is not that AI should never be used; it is that automation should not obscure who defines success, checks the work, and owns the decision.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errors
