Practical AI guidance
AI best practices: what public guidance says
A plain guide for team leads asking what best practices for AI actually are. Each practice below is tied to Anthropic's, GAO's or NIST's own page as read on 11 November 2026. The sorting is our wording.
AI best practices are the habits that public guidance returns to when organisations put AI to work. Three sources read on 11 November 2026 point the same way. Anthropic recommends finding the simplest solution possible and, for agents, testing extensively in sandboxed environments, and it describes agents pausing for human feedback at checkpoints. The U.S. Government Accountability Office organises its AI accountability framework around governance, data, performance and monitoring. NIST describes its AI Risk Management Framework as intended for voluntary use. No source gives one agreed list, so the grouping on this page is ours.
plain answer
AI best practices: where do they come from?
There is no single agreed list of best practices for AI. What exists is guidance from different bodies that overlaps. This page groups what three of them say, and the grouping is our wording.
The three are Anthropic's engineering article on building effective agents, published 19 December 2024, the GAO accountability framework GAO-21-519SP, dated June 30, 2021, and NIST's AI Risk Management Framework, released as version 1.0 on January 26, 2023.
Sources: Anthropic, Building effective agents (https://www.anthropic.com/engineering/building-effective-agents); GAO (https://www.gao.gov/products/gao-21-519sp); NIST (https://www.nist.gov/itl/ai-risk-management-framework), all read 11 November 2026.
plain answer
Best practices for AI: keep the design simple and test it
On design, Anthropic recommends finding the simplest solution possible and only increasing complexity when needed, and says to consider adding complexity only when it demonstrably improves outcomes. It says that for many applications, optimizing single LLM calls with retrieval and in-context examples is usually enough, and to add multi-step agentic systems only when simpler solutions fall short.
On testing, the same article recommends extensive testing in sandboxed environments, along with the appropriate guardrails, and says agents can pause for human feedback at checkpoints. For coding agents it says human review remains crucial.
Source: Anthropic, Building effective agents, read 11 November 2026.
plain answer
Organise accountability around four headings
GAO says its framework is intended to help managers ensure accountability and responsible use of artificial intelligence, and that it is organized around four complementary principles, which address governance, data, performance, and monitoring. For each principle it lists key practices.
A practical way to use that is as a checklist for any AI project: a named owner, data that suits the task, a measure of how well it performs, and a way to watch it after launch. The checklist wording is ours.
Source: GAO, GAO-21-519SP, read 11 November 2026.
plain answer
Use a risk framework, and check it is current
NIST describes the AI Risk Management Framework as a way to better manage risks to individuals, organizations, and society associated with artificial intelligence, and says it is intended for voluntary use. It also lists a Generative Artificial Intelligence Profile, NIST AI 600-1, released on July 26, 2024.
NIST's page says the AI RMF 1.0 is being revised as part of the White House AI Action Plan, so confirm the current version before you cite it in a policy.
Source: NIST, AI Risk Management Framework, read 11 November 2026.
security
What should a team write down first?
Write down five things. Who owns each AI system and answers for it. What data and tools it can reach. Which actions need a named person to approve them. How it was tested before launch. How it will be monitored and who reviews what the monitoring shows.
Those five answers cover the simple design, testing and accountability points above. They are our summary and not a formal standard.
kriv role
Where Kriv helps
Kriv AI helps teams with AI readiness and strategy and with governed deployment. For best practices, that means turning the points above into a short written standard for your team, with named owners and approval points. See our guide to an AI adoption strategy.
Kriv did not write the guidance cited on this page.
limits
What this page could not confirm
This page rests on the three pages listed below, as read on 11 November 2026. None presents a list called best practices, and we did not read the full GAO practices or NIST framework text. We quote no adoption rates, savings or other statistics because we did not find a source we could cite. Check the current pages before you plan.
Sources
Cited sources
Straight answers
Frequently asked questions about AI best practices: what public guidance says
What are AI best practices?
Habits that public guidance returns to: keep the design simple, test before launch, let a person review at checkpoints, and organise accountability around governance, data, performance and monitoring. No source gives one agreed list (read 11 November 2026).
Where should a team start with best practices for AI?
Anthropic recommends the simplest solution possible and adding complexity only when needed (read 11 November 2026).
Does GAO publish an AI framework?
Yes. GAO-21-519SP, dated June 30, 2021, is organized around governance, data, performance, and monitoring (read 11 November 2026).
Is NIST's AI framework required?
NIST says it is intended for voluntary use (read 11 November 2026).
Can Kriv help us write our AI standards?
Kriv AI helps teams turn public guidance into a short written standard with named owners and approval points.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call