Loop Engineering: A Smarter Way to Use AI
Most Malaysian business owners I talk to use AI the same way they used Google in 2015. Type a prompt, read the answer, type the next prompt. Repeat until the output might be usable. A newer pattern called loop engineering replaces that manual back-and-forth with an autonomous loop: you set a goal, define success criteria, and let the AI iterate until it meets the bar.
In short: Loop engineering lets you stop manually correcting AI outputs. You define the objective once, and the AI runs experiments, checks its own work, and repeats until it succeeds. Less prompting, better results, more sleep.
The Three Components of a Reliable Loop
What separates a real loop from a chatbot on repeat? Every reliable loop has three parts, as outlined in the original guide:
- Verifier – grades each attempt. This can be a passing test, a moving metric, or a build. Without a verifier, the agent simply agrees with itself on repeat.
- State – records what was tried, what failed, and what remains. A small side file lets the next run resume instead of restarting from scratch.
- Stop condition – prevents runaway cost. The loop halts when the goal is met, or after a set number of attempts.
These three pieces are not theoretical. They appear in every working loop, from simple content checks to advanced machine‑learning experiments.
Real Results and a Side‑by‑Side Comparison
The most famous example is the “Karpathy Loop.” In March 2026, Andrej Karpathy released an open‑source project called autoresearch. It hit nearly 90,000 GitHub stars within days. The concept is deliberately small: an agent edits only the training code, a verifier runs for five minutes per experiment, and the loop keeps or rolls back each change.
Karpathy pointed the loop at his already‑optimised GPT‑2 training code. Over two days it completed about 700 experiments, keeping 20 genuine fixes. Stacked together, those improvements cut training time by 11% (from 2.02 to 1.80 hours). Shopify CEO Tobi Lütke ran a similar loop overnight and reported a 19% improvement after just 37 experiments.
“If you have an objective metric, you are the bottleneck.” — Andrej Karpathy, via the loop engineering guide
Researchers have since built a loop on top of the loop: Bilevel Autoresearch. The outer loop watches the inner loop’s behavior and writes new mechanisms to break stale patterns. On the same benchmark, the bilevel version dropped the error metric five times more than the single loop (−0.045 vs −0.009). The big takeaway? The architecture matters more than a smarter model.
Here’s how the three approaches stack up at a glance:
| Aspect | Manual Prompting | Simple Loop (autoresearch) | Bilevel Loop |
|---|---|---|---|
| You define | Each step | The goal, once | The goal, once |
| Who iterates | You | Inner agent | Inner + outer agent |
| Verifier | You, manually | prepare.py (metric) |
Same metric, two levels |
| State | Chat only | Experiment log | Log + injected code |
| Human role | Engine | Author of instructions | Author of instructions |
The Bigger Picture: Loops for Your Business
You don’t need a GPU cluster to benefit from loop engineering. The same pattern works for any task that has a clear, measurable success criterion:
- Content creation – write a company blog post, then have the AI rewrite it until readability scores, keyword density, and tone each hit 8/10.
- Customer service – draft auto‑replies and let the loop test each variant against a satisfaction metric before sending.
- Data cleaning – transform messy sales spreadsheets until all schema checks pass — no manual correction required.
- Email sequences – generate and refine drip campaigns until open‑rate predictions exceed a threshold.
The common thread is an automatic gate that can fail the work. When that gate exists, you can hand the iteration to a loop and step away.
Build Your First Loop Today
You can feel the mechanic right now without any special tools. Copy the prompt below into your favourite AI model (ChatGPT, Claude, Gemini) and watch it self‑correct:
You will work in a loop until the task meets the bar.
TASK:
[describe exactly what you want produced]
SUCCESS CRITERIA (be strict):
- [criterion 1]
- [criterion 2]
- [criterion 3]
LOOP PROTOCOL, repeat every turn:
1. PLAN - state the single next step.
2. DO - produce or improve the work.
3. VERIFY - score the result 1-10 on each criterion. Be honest.
4. DECIDE - if every criterion is 8+, print FINAL and stop.
Otherwise print ITERATING and fix the weakest point first.
RULES:
- Never call it done until every criterion is 8 or higher.
- Each pass must fix the weakest score from the last VERIFY.
- Do not ask questions. Make a sensible assumption and continue.
Begin.Run this for your next email draft, product description, or internal memo. It will likely give you a better result than a one‑shot prompt — and it won’t ask for your input until it’s done.
If you’d like help setting up a loop for your specific business process, I’d be glad to walk you through it.
Book a free 15‑min call to see how loop engineering applies to your business →