Let the AI Do the Experimenting

in a scenario the place you might have loads of concepts on how one can enhance your product, however no time to check all of them? I guess you might have.

What if I instructed you that you simply not must do all of it by yourself, you possibly can delegate it to AI. It could run dozens (and even a whole lot) of experiments for you, discard concepts that don’t work, and iterate on those that truly transfer the needle.

Sounds superb. And that’s precisely the concept behind autoresearch, the place an LLM operates in a loop, repeatedly experimenting, measuring influence, and iterating from there. The strategy sounded compelling, and plenty of of my colleagues have already seen advantages from it. So I made a decision to attempt it out myself.

For this, I picked a sensible analytical process: advertising funds optimisation with a bunch of constraints. Let’s see whether or not an autonomous loop can attain the identical outcomes as we did.

Background

Let’s begin with some background to set the context. Autoresearch was developed by Andrej Karpathy. As he wrote in his repository:

In the future, frontier AI analysis was finished by meat computer systems in between consuming, sleeping, having different enjoyable, and synchronizing every now and then utilizing sound wave interconnect within the ritual of “group assembly”. That period is lengthy gone. Analysis is now completely the area of autonomous swarms of AI brokers working throughout compute cluster megastructures within the skies. The brokers declare that we are actually within the 10,205th technology of the code base, in any case nobody may inform if that’s proper or unsuitable because the “code” is now a self-modifying binary that has grown past human comprehension. This repo is the story of the way it all started. -@karpathy, March 2026.

The thought behind autoresearch is to let an LLM function by itself in an surroundings the place it may well repeatedly run experiments. It modifications the code, trains the mannequin, evaluates whether or not efficiency improves, after which both retains or discards every change earlier than repeating the loop. Finally, you come again and (hopefully) discover a higher mannequin than you began with. Utilizing this strategy, Andrej was in a position to considerably enhance nanochat.

The unique implementation was targeted on optimising an ML mannequin. Nevertheless, simialr strategy might be utilized to any process with a transparent goal (from decreasing web site load time to minimising errors when scraping with Playwright). Shopify later open-sourced an extension of the unique autoresearch, pi-autoresearch. It builds on pi, a minimal open-source terminal coding harness.

It follows an identical loop to the unique autoresearch, with a couple of key steps:

• Outline the metric you wish to enhance, together with any constraints.
• Measure the baseline.
• Speculation testing: in every iteration, the agent proposes an thought, writes it down, and exams it. There are three attainable outcomes: it doesn’t work (discard), it worsens the metric (discard), or it improves the goal (preserve it and iterate from there).
• Repeat: the loop continues till you cease it, enhancements plateau, or it reaches a predefined iteration restrict.

So the core thought is to outline a transparent goal and let the agent attempt daring concepts and study from them. This strategy can uncover potential enhancements to your KPIs by testing concepts your staff merely by no means had the time to discover. It positively sounds fascinating, so let’s attempt it out.

Activity

I want to take a look at this strategy on an analytical process, since in analytical day-to-day duties we regularly have clear targets and must iterate a number of instances to achieve an optimum answer. So, I went by way of all of the posts I’ve written for In direction of Information Science through the years and located a process round optimising advertising campaigns, which we mentioned within the article “Linear Optimisations in Product Analytics”.

The duty is kind of frequent. Think about you’re employed as a advertising analyst and must plan advertising actions for the subsequent month. Your aim is to maximise income inside a restricted advertising funds ($30M).

You have got a set of potential advertising campaigns, together with projections for every of them. For every marketing campaign, we all know the next:

• nation and advertising channel,
• marketing_spending — funding required for this exercise,
• income — anticipated income from acquired clients over the subsequent 12 months (our goal metric).

We even have some further data, such because the variety of acquired customers and the variety of buyer help contacts. We’ll use these to iterate on the preliminary process and make it progressively tougher by including additional constraints.

It’s helpful to present the agent a baseline strategy so it has one thing to begin from. So, let’s put it collectively. One easy answer for this optimisation is to concentrate on the top-performing segments by income per greenback spent. We will kind all campaigns by this metric and choose those that match inside the funds. After all, this strategy is kind of naive and may positively be improved, but it surely supplies place to begin. 

import pandas as pd

df = pd.read_csv('marketing_campaign_estimations.csv', sep='t')

# --- Baseline: grasping by revenue-per-dollar ---
df['revenue_per_spend'] = df.income / df.marketing_spending
df = df.sort_values('revenue_per_spend', ascending=False)
df['spend_cumulative'] = df.marketing_spending.cumsum()
selected_df = df[df.spend_cumulative <= 30_000_000]

total_spend = selected_df.marketing_spending.sum()
revenue_millions = selected_df.income.sum() / 1_000_000

assert total_spend <= 30_000_000, f"Finances violated: {total_spend}"

print(f"METRIC revenue_millions={revenue_millions:.4f}")
print(f"Segments={len(selected_df)} spend={total_spend/1e6:.2f}M")

I put this code in optimise.py within the repository. 

If we run the baseline, we see that the ensuing income is 107.9M USD, whereas the overall spend is 29.2M.

python3 optimise.py
# METRIC revenue_millions=107.9158
# Segments=48 spend=29.23M

Organising

Earlier than shifting on to the precise experiment, we first want to put in pi_autoresearch. We begin by organising pi itself by following the directions from pi.dev. Fortunately, it may be put in with a single command, supplying you with a pi coding harness up and working regionally which you can already use to assist with coding duties.

npm set up -g @mariozechner/pi-coding-agent # set up pi
pi # begin pi
/login  # choose supplier and specify APIKey

Nevertheless, as talked about earlier, our aim is to attempt the pi-autoresearch extension on high of pi, so let’s set up that as properly.

pi set up https://github.com/davebcn87/pi-autoresearch

I additionally wished some guardrails in place, so I created an autoresearch.config.json file within the root of my repo to outline the utmost variety of iterations. This helps restrict what number of iterations the agent can run and, in flip, retains token prices underneath management throughout experiments. You can even set a per-API-key spending restrict along with your LLM supplier for even tighter management.

{
  "maxIterations": 30
}

You’ll find all the small print on configuration in the docs.

That’s it. The setup is completed, and we’re prepared to begin the experiment.

Experiments

Lastly, it’s time to begin utilizing the autoresearch strategy to determine which advertising campaigns we should always run. I’m fairly certain our preliminary strategy isn’t optimum, so let’s see whether or not autoresearch can enhance it. Let the journey start.

I began autoresearch by calling the talent.

/talent:autoresearch-create

After that, autoresearch tries to deduce the optimisation aim, and if it fails, it asks for added particulars.

In my case, it merely inspected the code we carried out in optimise.py and created an autoresearch.md file summarising the duty. Right here’s what we obtained (a reasonably stable abstract, contemplating it solely noticed our baseline optimisation perform). We will see that it clearly outlined the metrics and constraints. I additionally preferred that it explicitly highlighted that altering the enter knowledge isn’t allowed. That’s guardrail.

# Autoresearch: maximize advertising marketing campaign income underneath funds

## Goal
Enhance `optimise.py` so it selects a set of marketing campaign segments with **most complete income** whereas respecting the mounted advertising funds of **30,000,000**. The present implementation is a grasping heuristic: it kinds by revenue-per-spend, takes a cumulative prefix, and stops as soon as the subsequent merchandise would exceed funds. Which means it may well depart funds unused and by no means contemplate cheaper worthwhile objects later within the sorted checklist.

The workload is tiny (62 rows), so higher-quality combinatorial optimization methods are possible sensible. We should always favor precise or near-exact choice logic over fragile heuristics when the runtime stays quick.

## Metrics
- **Main**: `revenue_millions` (tens of millions, larger is best) - complete chosen income divided by 1,000,000
- **Secondary**:
  - `spend_millions` - complete chosen spend divided by 1,000,000
  - `budget_slack_millions` - unused funds in tens of millions
  - `segment_count` - variety of chosen segments

## Find out how to Run
`./autoresearch.sh` - runs a fast syntax pre-check, then `optimise.py`, which should emit `METRIC title=quantity` strains.

## Recordsdata in Scope
- `optimise.py` - campaign-selection logic and metric output
- `autoresearch.sh` - benchmark harness and pre-checks
- `autoresearch.md` - session reminiscence / findings
- `autoresearch.concepts.md` - backlog for promising deferred concepts

## Off Limits
- `marketing_campaign_estimations.csv` - enter knowledge; don't edit
- Git historical past / department construction outdoors the autoresearch workflow

## Constraints
- Should preserve spend `<= 30_000_000`
- Should preserve the script runnable with `python3 optimise.py`
- No dataset modifications
- Maintain the answer easy and explainable until additional complexity yields materially higher income
- Runtime ought to stay quick sufficient for a lot of autoresearch iterations

## What's Been Tried
- Baseline code kinds by `income / marketing_spending`, computes cumulative spend, and retains solely the sorted prefix underneath funds.

After defining the duty, it instantly began the loop. It could run for a while, however you continue to retain visibility. You’ll be able to see each its reasoning and a few key stats within the widget (corresponding to the present iteration, greatest goal worth, and enchancment over the baseline), which is kind of useful.

Because it iterates, it additionally writes an autoresearch.jsonl file with full particulars of every experiment and the ensuing goal metric. This log could be very helpful each for reviewing what has been tried and for the mannequin itself to maintain monitor of which hypotheses it has already examined.

In my case, regardless of the configured restrict of 30 iterations, it determined to cease after simply 5. The agent explored a number of totally different methods: precise knapsack optimisation, search-space pruning, and a Pareto-frontier dynamic programming strategy. Let’s undergo the small print:

• Iteration 1: Reproduced our baseline strategy. The prefix-greedy technique (income/spend) reached 107.9M, however stopped early when objects didn’t match, lacking higher downstream mixtures. No breakthrough right here, only a sanity test of the baseline.
• Iteration 2: Precise knapsack solver. The agent switched to a branch-and-bound (0/1 knapsack) strategy and reached 110.16M income (+2.25M uplift), which is a transparent enchancment. A robust achieve already within the second iteration.
• Iteration 3: Dominance pruning. This iteration tried to shrink the search area by eradicating pairwise dominated segments (i.e., segments worse in each spend and income than one other). Whereas intuitive, this assumption doesn’t maintain within the 0/1 knapsack setting: a “dominating” section might already be chosen, whereas a “dominated” one can nonetheless be helpful together with others. In consequence, this strategy failed and dropped to 95.9M income, and was discarded. A very good instance of trial and error. We examined it, it didn’t work, and we instantly moved on.
• Iteration 4: Dynamic programming frontier. The agent switched to a Pareto-frontier dynamic programming strategy, but it surely achieved the identical consequence as iteration 2. From an analyst perspective, that is nonetheless helpful. It confirms we’ve possible reached the optimum.
• Iteration 5: Integer accounting. This iteration transformed all financial values from floats to integer cents to enhance numerical stability and reproducibility, however once more produced the identical remaining worth. It is sensible that the agent stopped there.

So ultimately, the optimum answer was already discovered within the second iteration and it matches the answer we present in my article with linear programming. The agent nonetheless tried a couple of different concepts, however stored ending up with the identical consequence and ultimately stopped (as an alternative of burning much more tokens).

Now we will end the analysis by working the /talent:autoresearch-finalize command, which commits and pushes every thing to GitHub. In consequence, it created a brand new department with a PR, saving each the modifications to the optimise.py code and the intermediate reasoning recordsdata. This manner, we will simply monitor what occurred all through the method.

The agent simply solved our preliminary process. Subsequent, let’s attempt making it extra real looking by including further constraints from the Operations staff. Assume we realised that we additionally want to make sure there are not more than 5K incremental buyer help tickets (so the Ops staff can deal with the load), and that the general buyer contact charge stays under 4.2%, since that is one among our system well being checks. This makes the issue tougher, because it provides additional constraints and forces the agent to revisit the answer area and seek for a brand new optimum.

To kick this off, I merely restarted the /talent:autoresearch-create course of, offering the extra constraints.

/talent:autoresearch-create I've further constraints for our CS contacts to make sure that our Operations
staff can deal with the demand in a wholesome means:
- The variety of further CS contacts ≤ 5K
- Contact charge (CS contacts/customers) ≤ 0.042

This time, it picked up precisely the place we left off. It already had full context from the earlier run, together with every thing we had finished to this point. On account of updating the duty, the agent revised the autoresearch.md file to incorporate the brand new constraints.

## Constraints
- Should preserve spend `<= 30_000_000`
- Should preserve further CS contacts `<= 5_000`
- Should preserve contact charge `<= 0.042`
- Should preserve the script runnable with `python3 optimise.py`
- No dataset modifications
- Maintain the answer easy and explainable until additional complexity yields materially higher income
- Runtime ought to stay quick sufficient for a lot of autoresearch iterations

It ran 8 further iterations and converged to the next answer (once more matching what we had seen beforehand):

• Income: $109.87M,
• Finances spent: $29.9981M (underneath $30M),
• Buyer help contacts: 3,218 (underneath 5K),
• Contact charge: 0.038 (underneath 0.042).

After introducing the brand new constraints, the agent reformulated the issue and switched to an precise MILP solver. It shortly discovered the optimum answer, reaching 109.87M income whereas satisfying all constraints. Many of the later iterations didn’t actually change the consequence, they only cleaned issues up: eliminated fallback logic, diminished dependencies, and improved runtime. So, as soon as the issue was well-defined, the agent stopped “looking out” and began “engineering”. What’s much more fascinating is that it knew when to cease optimising and didn’t run all the way in which to the 30-iteration restrict.

Lastly, I requested the agent to finalise the analysis. This time, for some cause, /talent:autoresearch-finalize didn’t push all of the modifications, so I needed to manually ask pi to create two PRs: one with clear code modifications, and one other with the reasoning and supporting recordsdata. You’ll be able to undergo the PRs if you wish to see extra particulars about what the agent tried.

That’s all for the experiments. We obtained superb outcomes and was in a position to see the capabilities of autoresearch. So, it’s time to wrap it up.

Abstract

That was a very fascinating experiment. The agent was in a position to attain the identical optimum answer we beforehand discovered, fully by itself. Whereas it didn’t push the consequence additional (which isn’t shocking given how well-studied issues like knapsack are), it was spectacular to see how an LLM can iteratively discover options and converge to a stable end result with out handbook steering.

I imagine this strategy has robust potential throughout a number of domains (from coaching ML fashions and fixing analytical duties to extra engineering-heavy issues like optimising system efficiency or loading instances). In lots of groups, we merely don’t have the time to check all attainable concepts, or we dismiss a few of them too early. An autonomous loop like this may systematically attempt totally different approaches and validate them with precise metrics.

On the similar time, that is positively not a silver bullet. There shall be circumstances the place the agent finds “optimum” options that aren’t possible in apply, for instance, enhancing web site loading velocity at the price of breaking person expertise. That’s the place human supervision turns into important: not simply to validate outcomes, however to make sure the answer is sensible holistically.

From what I’ve seen, this strategy works greatest when you might have a transparent goal, well-defined constraints, and one thing measurable to optimise. It’s a lot tougher to use it to extra ambiguous issues, like making a product extra user-friendly, the place success is much less clearly outlined.

Total, I’d positively advocate attempting out pi-autoresearch or comparable instruments by yourself issues. It’s a robust strategy to take a look at concepts you wouldn’t usually have time to discover and see what really works in apply. And there’s one thing nearly magical about your product enhancing when you sleep.

Disclaimer: I work at Shopify, however this put up is impartial of my work there and displays my private views.