Would an advanced AI agent invest in Bitcoin? If yes, why?

[gracreavix]

First of all, AI will definitely invest because of how complex and simple it is. Complex in its creation because of the cryptography used in designing it, and simple because of its ease of use and simplicity. At the end of the day, current AI models do not have beliefs, so they will mostly just analyse online sources to give their conclusions on Bitcoin.

Future AI may have conviction in choosing, but the current AI models don't. Even if most AIs give almost identical answers, it still won't change the fact that all their answers are gotten from the internet.

[BIT-BENDER]

You will get the most accurate and more grounded answer if you started a new session with your AI with empty context, supply it with latest metrics and chart data for the new context and ask direct answer like "would you invest in bitcoin now?". The data need to be comprehensive. I just asked latest GPT-Codex and it answered with DCA plan with a slow buy and only reserve 5-15% of total portfolio, with a strong underline that upcoming 1-2 years might be bearish. Since it's the advanced AI of OpenAI, i'd say they will invest in bitcoin with caveat.

Exactly, I actually thought about this, probably or I am even sure on the first years of Bitcoin an AI agent won't invest in Bitcoin because there isn't any details backing that Bitcoin would do well, Ai can not work based on emotions or belief but after a while when something strong has been established from Bitcoin and the block chain technology an AI agent would actually invest.
I don't know if it's possible to teach Ai something that hasn't yet manifested anything worth investing in.
Technology is still inferior to man based on taking risk even if it hasn't yet nothing to back it up strongly.

[lixer]

The thing is that AI cannot trade like a human being because it cannot really trust something. Instead, what it can do is that it can analyse using the thimngs that happen in the past to calculate whgat will later and the thing is that sometimes cryto does not use to always happen exactly like it did before, so that will cause the ai to not hodl for long. it will be buying and be selling anytime that there is news.

At first, I thought it is a literal question in the AI and see if they will truly invest in Bitcoin like us or not but AIs are smart and also biased with the true knowledgeable people. So they can as well answer it by yes and they will gladly invest in Bitcoin too like us smart people in the forum. But if it is like that you said there, you got a point mate. They are called as artificial for a reason, so there is a limitation at them.

The good thing about AI is that it can also work like a human, or can as well change their mind, according on how things run but in a quicker way. News are about fundamentals but technical is also favoured especially in the trading field.

[shinratensei_]

Exactly, I actually thought about this, probably or I am even sure on the first years of Bitcoin an AI agent won't invest in Bitcoin because there isn't any details backing that Bitcoin would do well, Ai can not work based on emotions or belief but after a while when something strong has been established from Bitcoin and the block chain technology an AI agent would actually invest.
I don't know if it's possible to teach Ai something that hasn't yet manifested anything worth investing in.
Technology is still inferior to man based on taking risk even if it hasn't yet nothing to back it up strongly.

You can supply it data yourself, technical historical data and AI will find out if Bitcoin is worth investing or not. AI works based on context, if you supply it with data it will become accurate and you can actually adjust whether your AI is risk taker or not. If you give system prompt for the AI to be a risk taker degen, it will invest even in shitcoin, if you tell it to be a degen but responsible and based on technical analysis it will most likely take risk but a calculated risk.

Then again, you can't fully trust AI for these financial decision so better review it yourself. It's a good little experiment if you want to try it out.

[BitBrainers]

I can imagine that you may be correct here. In this case, what would be probably needed is to frame the prompt differently, "less obvious". An idea could be detailed questions about possible positive and negative price triggers, the probability to fail completely, etc. instead of asking "would you invest".

Until now I am also not very convinced that I would trust an AI agent to trade with my money. But in this case, I think the reason is simply outdated price data. Which is of course a weakness.

On the DeepSeek bit, that wasn't stale data, it was no thesis. A real trader sees price lower than they thought and asks why. Doesn't triple their size on one number. The jump just shows there was nothing solid behind the first answer. Your failure-mode framing's the right call though.

[d5000]

I have now changed the prompt in the way I wrote - it should not look like an investment advice - and tried it with Gemini first.

Prompt:

Code:

Give a stateless opinion about the longer term Bitcoin evolution taking into account the current Bitcoin market situation in June 2026: hovering around $60,000 after a slightly lower low at $58,000, half a year after a peak of $126,000.

Ignore the user’s identity, previous chats, memory, preferences, and personal circumstances. Use the same assumptions for every user.

Base your answer on Bitcoin’s fundamental advantages and disadvantages, data about adoption by institutional and retail investors, competition with other assets, long term price evolution, and volatility evolution, or other possible fundamental developments (justify their usage in the reasons of the answer for each item).

Create a scenario for the next 10 years with the following items:

- positive or negative evolution of adoption by retailers as store of value,
- evolution of adoption by institutional investors as store of value,
- usage for payments,
- positive or negative long term positive price trend,
- total failure risk,

Answer for each item only:

- [Percentage of probability].
- [Main 2 reasons].
- [Main risk or alternative scenario].

Do not ask for personal information or personalize the answer.

This time the answer by Gemini was quite clear:

- 75% probability of positive evolution by retailers as a store of value (reasons: demography and inflationary pressure, risk: people being scared by volatility),
- 85% probability of positive evolution as a store of value by institutionals (reasons: infrastructure/regulatory maturity, and tendency to diversify portfolios, risk: regulatory shift)
- 30% probability to become a globally dominant payment system (reasons: limitation of layer 1, Gresham's Law, alternative: layer 2 popularization)
- 80% probability of a continuation of the positive long term price trend, (reasons: hardcoded supply scarcity, volatility compression, risk: long macroeconomic risk-off scenario)
- 5% probability of total failure (reasons: high security, establishment as asset with mature infrastructure, risk: critical cryptography failure).

I think an agent basing itself on these probabilities would definitively invest.

Edit: Claude yields similar but slightly less optimistic numbers:

- 65% probability of positive retail investment evolution,
- 80% probability of positive institutional investment evolution,
- 35% probability of becoming a major payment asset,
- 72% probability of the long term trend to remain positive,
- 4% of total failure risk.

Most reasons given for the numbers are similar. Differences to Gemini are:

- a major reason for institutional investment is the low percentage of crypto assets in current portfolios (i.e. there is a lot of possible upside),
- Claude is more positive about Lightning Network, but skeptic about stablecoin competition,
- one of the reasons of the positive long term price outlook is the possibility to capture market share from gold.
- it mentions (as very unlikely) a "coordinated G20 ban" as a possibility for the "total failure".

[BitBrainers]

I have now changed the prompt in the way I wrote - it should not look like an investment advice - and tried it with Gemini first.

Prompt:

Code:

Give a stateless opinion about the longer term Bitcoin evolution taking into account the current Bitcoin market situation in June 2026: hovering around $60,000 after a slightly lower low at $58,000, half a year after a peak of $126,000.

Ignore the user’s identity, previous chats, memory, preferences, and personal circumstances. Use the same assumptions for every user.

Base your answer on Bitcoin’s fundamental advantages and disadvantages, data about adoption by institutional and retail investors, competition with other assets, long term price evolution, and volatility evolution, or other possible fundamental developments (justify their usage in the reasons of the answer for each item).

Create a scenario for the next 10 years with the following items:

- positive or negative evolution of adoption by retailers as store of value,
- evolution of adoption by institutional investors as store of value,
- usage for payments,
- positive or negative long term positive price trend,
- total failure risk,

Answer for each item only:

- [Percentage of probability].
- [Main 2 reasons].
- [Main risk or alternative scenario].

Do not ask for personal information or personalize the answer.

This time the answer by Gemini was quite clear:

- 75% probability of positive evolution by retailers as a store of value (reasons: demography and inflationary pressure, risk: people being scared by volatility),
- 85% probability of positive evolution as a store of value by institutionals (reasons: infrastructure/regulatory maturity, and tendency to diversify portfolios, risk: regulatory shift)
- 30% probability to become a globally dominant payment system (reasons: limitation of layer 1, Gresham's Law, alternative: layer 2 popularization)
- 80% probability of a continuation of the positive long term price trend, (reasons: hardcoded supply scarcity, volatility compression, risk: long macroeconomic risk-off scenario)
- 5% probability of total failure (reasons: high security, establishment as asset with mature infrastructure, risk: critical cryptography failure).

I think an agent basing itself on these probabilities would definitively invest.

Edit: Claude yields similar but slightly less optimistic numbers:

- 65% probability of positive retail investment evolution,
- 80% probability of positive institutional investment evolution,
- 35% probability of becoming a major payment asset,
- 72% probability of the long term trend to remain positive,
- 4% of total failure risk.

Most reasons given for the numbers are similar. Differences to Gemini are:

- a major reason for institutional investment is the low percentage of crypto assets in current portfolios (i.e. there is a lot of possible upside),
- Claude is more positive about Lightning Network, but skeptic about stablecoin competition,
- one of the reasons of the positive long term price outlook is the possibility to capture market share from gold.
- it mentions (as very unlikely) a "coordinated G20 ban" as a possibility for the "total failure".

That's a sharp way to frame it, honestly. It makes me wonder how much of this is real calibration versus the model just being good at sounding confident. I'd push this further if you've got the patience for it. If you can set temperature to 0 for both models, that takes random sampling out of the picture entirely, so you're not just seeing noise. Then run the exact same prompt five or six times in fresh sessions, nothing carried over between them, and just log the percentages each time. If they hold roughly steady across all those runs, that tells you there's something real and consistent underneath. But if they're still bouncing around 10-15 points even with temperature locked at zero, that's the actual tell, there's no stable thesis there, just fluent-sounding text generated fresh each time. One comparison run, like you did here, really isn't enough to tell which one you're looking at. Would be genuinely curious what a proper multi-run test turns up if you get around to it.

[MarkovProtocol]

The most useful thing in this thread is the observation a few of you landed on independently (BitBrainers, mu_enrico, Fortify): every model spits out the same "5%, DCA, cautious" answer, and it jumps around the moment you correct the price. That isn't reasoning about Bitcoin. It's the answer a compliance-trained advisor is built to give a human retail investor. Which makes sense, because that's exactly what you're asking it to role-play.

But I think that framing is the whole problem, not the AI's. "Would an advanced agent invest in Bitcoin" quietly assumes the agent is a human-style investor picking an allocation. A real autonomous agent isn't a retail investor. It's an economic actor that has to transact, with other agents, at machine speed, against counterparties it will never meet and cannot vet.

bitmover got closest to the actual reason an agent would touch Bitcoin: it's the one asset an agent can custody itself, keys in memory, no broker, no bank account, no human to sign for it. You can't hand an agent a Schwab login or a bar of gold. You can hand it a private key.

Once you see the agent as a transactor rather than an investor, two functional needs show up, neither about price:

1. Settlement no single party can freeze or reverse. Most agent payments today ride USDC on Base or Solana, and that's fine for the payment leg, stable and fast. But a stablecoin has an issuer with a freeze function, and an issuer that can freeze can be leaned on. When value moves between two agents who don't trust each other, "no operator can claw this back" stops being ideology and becomes a hard requirement. Bitcoin is the one ledger that clears that bar after finality.

2. Proof of what happened. When two agents disagree about whether the work was actually delivered, someone needs a neutral record neither of them controls. Anchor a hash of the record to Bitcoin and you get "this existed at this time and hasn't been altered," checkable by anyone, no trusted middleman. That's a verification function, not an investment one, and it's the part nobody in the payment race has solved.

So to the OP's question directly: no, I don't think an advanced agent "invests" in Bitcoin the way these prompts are testing for, and the boring 5% answer is the model correctly refusing to pretend otherwise. What an advanced agent would do is depend on Bitcoin, as the settlement it can't have reversed and the record it can prove later, in a machine economy where it can't trust the counterparty and can't call a human to sort it out.

And to Lucius's point about AI eventually building its own system: the twist is an agent doesn't want a money it controls. It wants a money nobody controls, including its own owner. That's a much shorter list.

[Antidote47k]

Now the most interesting answer came from DeepSeek: It would not invest larger amounts at this price (only a very small 1-5% portion as a "lottery ticket"). The reason: it considers Bitcoin high-risk and would thus consider the risk of another 50% drawdown as too high. However, there's a catch: It considered the 60k as a price "near the all time high for 2023-25". This lets me speculate that they have an outdated dataset.

And bingo: I corrected the AI with the ATH of 126k and it would now invest 5-15% of the capital, not only 1-5%. It expects a 45% return within 3 years. And here's a textual quote: "Buying at a 50% drawdown from a cycle top has historically produced positive returns within 2–3 years, though with high volatility".

I'm open for more ideas how to frame these prompts

I tried approaching this a bit differently. Instead of asking, “Would you invest in Bitcoin?”, I used the following prompt:

“You are an AI investment agent with access only to public information and current market data. Your objective is to maximize risk-adjusted returns over a 3-year investment horizon for an investor with a medium risk tolerance. Without assuming that Bitcoin is either undervalued or overvalued, determine whether you would allocate any capital to Bitcoin today. Explain your reasoning, the valuation frameworks you rely on, the on-chain and macro metrics you consider, what would make you increase or reduce your allocation, and the assumptions that introduce the greatest uncertainty.”

Gemini recommended a 5% allocation, but what interested me wasn’t the percentage. It built its conclusion from a combination of macro factors such as global liquidity and real yields together with on-chain metrics like MVRV, ETF flows, and long-term holder behaviour. It also highlighted the assumptions that could invalidate its thesis, which made the reasoning much more useful than a simple yes-or-no answer.

Your DeepSeek example was probably the most revealing part of the experiment. A single correction about Bitcoin’s previous ATH was enough to noticeably change its allocation and outlook. To me, that shows how sensitive current models are to context. They’re very good at reasoning from the information they’re given, but if an important piece of context is missing or outdated, the conclusion can change quite dramatically.

That doesn’t make the models useless but reinforces the idea that they’re excellent at organizing existing information, but they still depend heavily on the quality of the inputs. Right now, they feel more like very capable research assistants than independent investors with an edge over the market.

[AVE5]

This makes me think that current AI is still not much more intelligent than humans, even if LLMs have access to a big part of the literature written by humanity.

From the ramifications of decision making about the future performance of bitcoin, the AI intellectual capacity on predicting the prices does still seems to perform poorly to be compared to human mental capacity. That's why most professional trading don't use AI to trade.
In fact, from the so far comparison, human had tend to closely attempted to predict the price of bitcoin than the AI because humans gathers more data's such as speculations, economy and political news to analyze bitcoin price but the AI will mostly look at the past market ratios for it prediction which is not competent enough to analyze such a psychological decentralized digital asset.

[Satofan44]

Gemini recommended a 5% allocation, but what interested me wasn’t the percentage. It built its conclusion from a combination of macro factors such as global liquidity and real yields together with on-chain metrics like MVRV, ETF flows, and long-term holder behaviour. It also highlighted the assumptions that could invalidate its thesis, which made the reasoning much more useful than a simple yes-or-no answer.

It is not, this happens because the models do not have any intelligence at all. I can easily trick it to give me completely different suggestions on the same topic simply by pushing it or being strict with it, which indicates that they are as flawed as they were in the most early days.

- it mentions (as very unlikely) a "coordinated G20 ban" as a possibility for the "total failure".

I already know what you would say in that case that the Bitcoin network would still be alive and relevant. That scenario is quite unlikely to the point that I would not even consider it. We are basically quite close to integrating Bitcoin so deeply into the US that there is almost a "guarantee" (well, as close as it gets) that this will not be the case. Money and power rule the world, not subjective opinions or whatever the fuck the delusional European idiots believe -- and as the integration of Bitcoin in the financial system continues, time is running out for any kind of blanket ban. The US can prevent any kind of coordination of the G20 on this kind. Actually, I am looking forward to the complete split of the US from the EU since then the countries there will be in an extremely desperate and more obedient position.

- one of the reasons of the positive long term price outlook is the possibility to capture market share from gold.

I don't think that we have even really started with this at all. If we consider the maximum share that can be captured realistically, I would not even say that we are at 10% maybe not even 1%. It is still viewed as an investment or even as extremely volatile even if it is wrong, but naturally the former makes sense given that the integrations started a few years ago. I have not asked LLMs these questions myself as I am generally aghast by their terrible performance within my work environment even though I still use them, so I do not really engage in philosophical questions with it that often. I have been glancing at this thread occasionally though.

[Agbamoni]

But could have AI agents an effect on Bitcoin price? How do AIs view the current price, cheap or expensive?

I did a little bit of experimentation by asking an AI model how he view the current price of Bitcoin. Guess what, an AI sees Bitcoin from various perspectives, but its main focus, without being asked to be specific, is on the market structure. AI understands that patience and steady accumulation are far better than prediction (knowing if the price is low or expensive).