Let the AI Do the Experimenting

in a state of affairs the place you will have loads of concepts on methods to enhance your product, however no time to check all of them? I wager you will have.

What if I advised you that you simply now not must do all of it by yourself, you’ll be able to delegate it to AI. It could actually run dozens (and even a whole bunch) of experiments for you, discard concepts that don’t work, and iterate on those that really transfer the needle.

Sounds wonderful. And that’s precisely the thought behind autoresearch, the place an LLM operates in a loop, constantly experimenting, measuring impression, and iterating from there. The method sounded compelling, and plenty of of my colleagues have already seen advantages from it. So I made a decision to strive it out myself.

For this, I picked a sensible analytical activity: advertising price range 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:

Sooner or later, frontier AI analysis was completed by meat computer systems in between consuming, sleeping, having different enjoyable, and synchronizing infrequently utilizing sound wave interconnect within the ritual of “group assembly”. That period is lengthy gone. Analysis is now fully the area of autonomous swarms of AI brokers operating throughout compute cluster megastructures within the skies. The brokers declare that we are actually within the 10,205th era of the code base, in any case nobody may inform if that’s proper or incorrect 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 concept behind autoresearch is to let an LLM function by itself in an atmosphere the place it could constantly run experiments. It adjustments the code, trains the mannequin, evaluates whether or not efficiency improves, after which both retains or discards every change earlier than repeating the loop. Ultimately, you come again and (hopefully) discover a higher mannequin than you began with. Utilizing this method, Andrej was in a position to considerably enhance nanochat.

The unique implementation was targeted on optimising an ML mannequin. Nevertheless, simialr method could be utilized to any activity 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 the same loop to the unique autoresearch, with a number of key steps:

• Outline the metric you need to enhance, together with any constraints.
• Measure the baseline.
• Speculation testing: in every iteration, the agent proposes an concept, writes it down, and assessments it. There are three doable outcomes: it doesn’t work (discard), it worsens the metric (discard), or it improves the goal (hold 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 concept is to outline a transparent goal and let the agent strive daring concepts and study from them. This method can uncover potential enhancements to your KPIs by testing concepts your crew merely by no means had the time to discover. It undoubtedly sounds fascinating, so let’s strive it out.

Job

I wish to take a look at this method on an analytical activity, since in analytical day-to-day duties we regularly have clear goals and must iterate a number of occasions to achieve an optimum answer. So, I went via all of the posts I’ve written for In the direction of Knowledge Science through the years and located a activity round optimising advertising campaigns, which we mentioned within the article “Linear Optimisations in Product Analytics”.

The duty is sort of widespread. Think about you’re employed as a advertising analyst and must plan advertising actions for the following month. Your aim is to maximise income inside a restricted advertising price range ($30M).

You will have 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 prospects over the following 12 months (our goal metric).

We even have some extra info, such because the variety of acquired customers and the variety of buyer help contacts. We’ll use these to iterate on the preliminary activity and make it progressively more difficult by including additional constraints.

It’s helpful to present the agent a baseline method so it has one thing to start out from. So, let’s put it collectively. One easy answer for this optimisation is to give attention to the top-performing segments by income per greenback spent. We are able to type all campaigns by this metric and choose those that match throughout the price range. In fact, this method is sort of naive and might undoubtedly be improved, but it surely supplies an excellent 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 operating regionally you could 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 strive the pi-autoresearch extension on prime of pi, so let’s set up that as nicely.

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

I additionally needed 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 below management throughout experiments. You too can set a per-API-key spending restrict together with your LLM supplier for even tighter management.

{
  "maxIterations": 30
}

You will discover all the main points on configuration in the docs.

That’s it. The setup is finished, and we’re prepared to start out the experiment.

Experiments

Lastly, it’s time to start out utilizing the autoresearch method to determine which advertising campaigns we must always run. I’m fairly certain our preliminary method 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 applied in optimise.py and created an autoresearch.md file summarising the duty. Right here’s what we received (a reasonably strong abstract, contemplating it solely noticed our baseline optimisation operate). We are able to 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 an excellent guardrail.

# Autoresearch: maximize advertising marketing campaign income below price range

## Goal
Enhance `optimise.py` so it selects a set of marketing campaign segments with **most complete income** whereas respecting the mounted advertising price range of **30,000,000**. The present implementation is a grasping heuristic: it types by revenue-per-spend, takes a cumulative prefix, and stops as soon as the following merchandise would exceed price range. Meaning it could go away price range unused and by no means take into account cheaper worthwhile objects later within the sorted record.

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` (thousands and thousands, 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 price range in thousands and thousands
  - `segment_count` - variety of chosen segments

## The best way 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 exterior the autoresearch workflow

## Constraints
- Should hold spend `<= 30_000_000`
- Should hold the script runnable with `python3 optimise.py`
- No dataset adjustments
- 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 types by `income / marketing_spending`, computes cumulative spend, and retains solely the sorted prefix below price range.

After defining the duty, it instantly began the loop. It could actually 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 (resembling the present iteration, finest goal worth, and enchancment over the baseline), which is sort 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 may be very helpful each for reviewing what has been tried and for the mannequin itself to maintain observe 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 method. Let’s undergo the main points:

• Iteration 1: Reproduced our baseline method. The prefix-greedy technique (income/spend) reached 107.9M, however stopped early when objects didn’t match, lacking higher downstream combos. No breakthrough right here, only a sanity examine of the baseline.
• Iteration 2: Actual knapsack solver. The agent switched to a branch-and-bound (0/1 knapsack) method and reached 110.16M income (+2.25M uplift), which is a transparent enchancment. A powerful 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 could already be chosen, whereas a “dominated” one can nonetheless be helpful together with others. Because of this, this method failed and dropped to 95.9M income, and was discarded. 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 method, but it surely achieved the identical outcome 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 ultimate 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 number of different concepts, however saved ending up with the identical outcome and finally stopped (as an alternative of burning much more tokens).

Now we are able to end the analysis by operating the /talent:autoresearch-finalize command, which commits and pushes every thing to GitHub. Because of this, it created a brand new department with a PR, saving each the adjustments to the optimise.py code and the intermediate reasoning information. This manner, we are able to simply observe what occurred all through the method.

The agent simply solved our preliminary activity. Subsequent, let’s strive making it extra practical by including extra constraints from the Operations crew. Assume we realised that we additionally want to make sure there are not more than 5K incremental buyer help tickets (so the Ops crew can deal with the load), and that the general buyer contact charge stays beneath 4.2%, since that is one in all our system well being checks. This makes the issue more difficult, 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 extra constraints for our CS contacts to make sure that our Operations
crew can deal with the demand in a wholesome means:
- The variety of extra 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 completed to this point. On account of updating the duty, the agent revised the autoresearch.md file to incorporate the brand new constraints.

## Constraints
- Should hold spend `<= 30_000_000`
- Should hold extra CS contacts `<= 5_000`
- Should hold contact charge `<= 0.042`
- Should hold the script runnable with `python3 optimise.py`
- No dataset adjustments
- 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 extra iterations and converged to the next answer (once more matching what we had seen beforehand):

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

After introducing the brand new constraints, the agent reformulated the issue and switched to an precise MILP solver. It rapidly discovered the optimum answer, reaching 109.87M income whereas satisfying all constraints. Many of the later iterations didn’t actually change the outcome, they simply cleaned issues up: eliminated fallback logic, lowered 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 motive, /talent:autoresearch-finalize didn’t push all of the adjustments, so I needed to manually ask pi to create two PRs: one with clear code adjustments, and one other with the reasoning and supporting information. 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 received wonderful outcomes and was in a position to see the capabilities of autoresearch. So, it’s time to wrap it up.

Abstract

That was a extremely fascinating experiment. The agent was in a position to attain the identical optimum answer we beforehand discovered, utterly by itself. Whereas it didn’t push the outcome 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 strong end result with out guide steerage.

I consider this method has sturdy 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 occasions). In lots of groups, we merely don’t have the time to check all doable concepts, or we dismiss a few of them too early. An autonomous loop like this may systematically strive totally different approaches and validate them with precise metrics.

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

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

General, I’d undoubtedly suggest attempting out pi-autoresearch or related instruments by yourself issues. It’s a robust approach to take a look at concepts you wouldn’t usually have time to discover and see what truly works in observe. And there’s one thing nearly magical about your product bettering when you sleep.

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