Deep Dive into Devin AI - A strategist’s analysis
As mentioned earlier this week, Devin AI is a new programming copilot specifically fine-tuned for software engineering tasks:
and iteratively work through programming challenges other engineers may find particularly troublesome:
As far as we can tell from the overview videos, this covers quite a few programming languages:
and an unknown number of clouds, but presumably building for integration into all of them with a presumed order of max adoption first.
More use cases can be found here:
How does this work?
Logging into the webui, it looks like we get a connection to a Ubuntu container including:
One of the things I love is Devin AI is relying on automated chain of thought reasoning without forcing the user to come up with an intermediary response.
For those just getting into the LLM world, chain of thought prompting is:
From there, you can direct Devin through the normal programming workflow like you would ChatGPT, yet be much more vague, provide links to a URL for background on the task, and get high quality results towards the end of the workflow.
Tongue-in-cheek, it does rely on a developer’s favorite debugging method: debug via println.?
Nothing can top the old classics and it seems the old favorite of students and engineers everywhere made it to production software!
That said, the overall versatility of the tool is amazing and will help a lot of engineers work incredibly efficiently.
Realities of the tooling
Obvious issues and tie-ins to past material
It’s obvious at first glance how this will impact jobs and company bottom-line, shrinking the overall market and making it more competitive while companies enjoy record-breaking profit.
That said, though, if we take into consideration information provided earlier this week, we know that super-exponential growth and new problems creep up, a la climate change for example, that we should have enough room to reorient ourselves within the professional environment while making a stable living.
Acknowledging my own fortunes, I’ve got a loving wife who supports us financially while I act as an entrepreneur and I want to say just how much I appreciate that.
That said, though, I want to help everyone take advantage of the natural rhythms in the economy and it likely means we will have to become entrepreneurs and business owners at some point in our life if we want to maintain the standard of living and build wealth.
For this to happen, though, the best thing to do is to start a business on the side safely while you have a job and do the whole buildup until it maintains a revenue stream equivalent to your salary for 6-12 months.
This is how you know it is safe to jump ship full-time.
Some employment contracts can be quite restrictive though so stay safe!
Business Realities
Since we have limited information from Cognition, we can assume the project is in a beta state as shown on the screen and that most engineers there are probably getting paid $150-250K as a fair market rate.
Given the $21M Series A round just recently raised by the company, we can tell they are ready to start scaling, but this sum of money may last 2 years at most.
领英推荐
Knowing VCs this isn’t an issue, but we can rest assured the company is under some key growth metrics given the sum of investment.
For those unaware, VCs syndicate investments through a big pool of money called a fund raised every 5-7 years, hence the 5-7 year timeline, under which they must show returns to investors (i.e. high net-work individuals).
Usually at the start, they’ll fund 10 companies or so, 5 will die immediately, 3-4 will reach a not-great stability, and 1-2 will unicorn which this one is easily likely to do.
Because of the expectations from the LPs (folks whose money is being invested), this company will have to reach a 10x multiple reaching $100M revenue and IPO or be sold off.
Realistically, for the money needed to fund this project, it may have to reach a 15-20x multiple to really be useful for all the shareholders to see a reasonable return on their investment.
Product realities
Unfortunately, we don’t know the full reach of the product currently since it just got out of stealth mode, but it seems fair to assume what you see is what you get at the moment.
This means we don’t know how many clouds this is integrated into, how the backend was built, nor what is needed to perform maintenance on, but knowing companies similar to this, they’ll be able to get up to 20-30 engineers easily, not to mention other critical roles like sales, marketing etc.
Regular maintenance of the product won’t be cheap though, so it’s easy to guess that all the varied deployment patterns, secure code scanning, live debugging features and more are currently on the development roadmap.
Keeping this in mind, we can perhaps see getting one cloud or two by the time they go public; maybe more if they use this tool to program itself, but paying for their own operating expenditures out of the same bucket, this will be interesting.
Taking a stab at the future
Financial guesswork
Hazarding a wild guess here, there will be a freemium level of support into a pricing structure that probably supports companies starting at upper small-cap businesses and lower mid-cap businesses.
Given the amount of money needed to fund this tool, I can’t see it being available to folks who haven’t reached appropriate economies of scale yet.
Perhaps it will be available for smaller businesses, but I don’t see that happening and definitely see a gap in the market for someone to fill in there.
This will probably be a higher market play perhaps with a niching down strategy on higher-end customers who can afford to help make the presumed stresses this would need to launch into the market.
Business guesswork
Assuming a higher end audience, they’ll need to get common enterprise features enabled sooner than later so it could be adopted by the target market.
It cross-cuts all industries which is helpful, but there will be a lot of money spent still trying to get those features implemented.
Perhaps judgment is still off based on the number of examples provided, but it still seems to be that it won't necessarily be as game-changing when we get 6-12 months down the road.
Plus, knowing competition with others shown in the benchmark who are much better resourced, Cognition’s lead probably won’t last long now that it has posted to the scoreboard.
Likely scenario here? This gets built up and sold into one of the bigger tech companies who can give it the level of support it needs to go all the way and integrate with the latest models.
Reinforcing the economic outlook
In the near term, we can expect a tempered response in my mind as the tool grows.
This is likely somewhere around the next 6-12 months for the first set of customers accounting for what we’re aware of needing done and then the next 2 years for really getting into a profit level as the business starts to scale.
Expectations are this will be priced for upper level small businesses and above. Others should expect to rely on ChatGPT and other tooling until they can get into a range where buying Devin would have an appropriate impact on the business.
This will leave the target market perhaps under-served with existing tooling or allow someone at lower cost to step in and produce a better tool.
Plugs
Blockchain Developer | Full Stack Web Developer | Proficient in JavaScript, Solidity, and Full Stack Technologies | Building Innovative Solutions for the Future
8 个月problem is how the Browser works with cors in it
Full Stack Web Developer ? Helping startup companies founder save time & money ? Helping coaches at software agencies.
8 个月Wow amazing ??. It's really ossam.