309 comments

[ 0.21 ms ] story [ 3589 ms ] thread
I never thought the day that IBM would make complete slop.

But it is IBM after all.

I bet this will be as successful as Microsoft Bob.
I used Bob extensively at an IBM acquisition, it was absolutely dreadful. Bobcoins.
This can only mean that IBM Clippy is next.
Nobody got fired for using IBM Bob
Pretty sure a lot of people got fired over Microsoft Bob though
Well, you could say Melinda got a big promotion for a while.
Nobody who got fired uses IBM Bob
Can we say it's Bob's fault, though? Bob did it, not me.
Nope, but they'll probably get laid off at some point to boost share price.
Should have been IBM Tom, after legendary IBM CEO Thomas J. Watson. But perhaps they didn't want to sully his name.
IBM already have IBM Watson which is probably named after the CEO
Well done on understanding the reference.
As somebody already mentioned, they already had a big product named after their CEO. IBM Watson was the next big thing after Deep Blue, and it went pretty viral after defeating Ken Jennings and Brad Rutter on Jeopardy back in ~2011.

It was a bit of a market play and of course has since been completely upstaged by the invention of LLMs, but it still definitely got them the attention they wanted at the time.

https://www.ibm.com/history/watson-jeopardy

Yes, I know, I got a job offer to work in the Watson research lab around that time, at the T.J. Watson facility. I declined because they weren’t remote-friendly, among other things.

But IBM Tom would at least have a rationale. WTF is IBM Bob and why does it have the same name as an old Microsoft product?

This is embarrassing. It would have been better to release nothing than this.
(comment deleted)
I like how the solitary review is word salad.
Totally reminds me of "HP Offers 'That Cloud Thing Everyone Is Talking About'"

https://youtube.com/watch?v=9ntPxdWAWq8

How you doing fellow kids?
yet hp exist after decades, just a terrible printer corp going long term is insane really
I suspect their enterprise computer business creates more revenue than printers.
They're the 2nd biggest PC manufacturer in the world. Sell more computers than Dell, Apple, etc.

Only beaten by Lenovo.

yet they gather PC parts right ? like did they create any board or ram or any great hardware, no ?
Yes, but as a consumer you want to pretend they aren't. Seriously, buy from anyone else.
HP was never just a printer corp. ;-)
HP is long gone. What's left is just a printer and PC company that didn't bother to change the sign on the building.

It's as if I bought an old telephone exchange building and called it Bell Labs.

You couldn't, mostly because Bell Labs is owned by Nokia now.
Interesting, TIL! I wonder what Nokia's market cap is... geez, still over $50B. Another surprise.
That's basically all Nokia(-Siemens) Networks.
It's Nokia again now, bought out Siemens' share more than a decade ago.
AT&T used to exist. Because it was a monopoly, it was broken up into the baby bells in 1982. One of which was Southwestern Bell, which eventually bought up all the other baby bells and what was left of AT&T. And then in 2005 it renamed itself...AT&T.
The fun parts of HP split off years ago into Agilent, Keysight, and HPE. Current HP is just a printer and office PC corp.
The most valuable offspring of HP is Broadcom, just as ASML is the most valuable offspring of Philips.
It kind of is - when I looked at it 6 months ago it was just a fork of vscode with old ass versions of Claude and the only unique things were java and ibm I mcp clients or servers or something built in. For those works I'd rather just have the skill or mcp for my harness of choice. I think they weren't allowed to use external AI so they run the models in house and remade cursor. Not bad to have competition but the name is awful - as others have said it's been used before by Microsoft in the 90s to make a friendly desktop giving us the winxp search dog and clippy.
(comment deleted)
I love this. Thanks for sharing!
This sounds a lot like Spotify's XIRP. They managed to vibe something useful for their own engineers, were super hyped, and decided to release it in public. However what are the odds this harness will stand the test of time against actual harness makers like OpenAI or Anthropic? Neither Spotify or IBM are focused on making models or harnesses, these are just side-products.
To be fair Codex and Claude Code don't have a lot of secret sauce either. This is not to say that the harness doesn't matter at all or that you can't do any better, but whenever they benchmark it always turns out that you don't lose that much with even a very minimal agent. If a different harness makes better sense for your workflow I don't see why you wouldn't switch (or make your own if you have the resources to do so).

I'm also sticking to first-party agents, but that's more because I feel no compelling reason to switch and CLI agents fit well with my spartan technology choices. YMMV.

> actual harness makers

I agree IBM Bob is not the revolution, but…

We are still in the punch card era of LLMs. Maybe even the Altair era.

None of this stuff has really been figured out yet.

IBM is training their own models (for instance IBM granite) as well as watsonx as a platform. They are trying to build a products for enterprise as well as their specific AI modernization models.
IBM simply does not have the capital to compete with training frontier models or otherwise.
Sure! IBM and Spotify, two companies united in being famous for not being software consulting businesses and definitely not ever developing software development tools.
Interesting they are renting out agent skills. You can pick up "Java Modernization" at $20 a month for "Specialist workflows to help with modernizing Java applications", or "IBM i" for "Specialist modes, workflows, and skills to work with IBM i applications".
markdown as a service is crazy
Is it that diff from code as a service? It’s human intent transcribed into something useful.
Code is precise. Human intent is anything but.
IBM hasn't made a product that works since Db2 so I am fairly confident this is a flaming pile of garbage.
No, you are wrong. IBMer here. I was at first agains it, my background is for 2 decades startup(s) before I joined IBM, and I started slowly using IBM Bob and I must say it is really good and helping. With many integrations that works :)
lol, I've used it and Bob is a dumpster fire.
Do you have to have experience with AWS Kiro? I'd be interested in a comparision between them. :)
rlly? cs it is IMHO based on anthropic / openai models. Perhaps they identified BFU and gave you some
so it tops out at sonnet 4.5 as the largest model it will use. the harness gives you no choice in the model so you have to hope it doesn't decide your task isn't complex enough and auto-route you to a smaller/weaker model which you then go round and round in circles with where a larger model would've 1 shotted it
What is good about it then? What model do you use? Which integration are we talking about? Is it open source? How much the "transparent pricing" "bobcoin" is worth?
It's not good, they hide the model (their whole thing is hidden model routing to keep costs low, but predictably yields bad results), there are no novel integrations not available in any other harness, it's closed source, and bobcoin is worthless because it's only usable with Bob.
And what? IDK about BOB that is available for IBM customers, but closed source? As someone wrote in different thread is based on VScode, calling BOB's API. Also do you know how chatgpt works? same for claude? Agents? Governance? You know nothing and now just hating here :)
the bob team aggressively blocks you from calling the bob api and is constantly finding new ways to thwart calling the api directly
It uses hidden models, mostly Claude and open ai models. They try to check the prompt and route to the model that is capable of it, thus saving money.
I think this is it ^^^ Routing and optimizations :)
Is it true that employees have to earn Bobcoins before they can use Bob?
You get more bobcoins by using your old allowance and running out, then submitting GitHub issues that get ignored.
Are you allowed to use Cursor, Codex, Claude, or OpenCode or is Bob all they allow?
I have been told only Bob is allowed but not 100% on that.
Also copilot and MS365 copilot, many local models etc... Many haters (companies, etc...) can only dream about availability of models we usually have here.
Official policy is to only use Bob. All of the best people I work with are using other tools because Bob is trash.
When people say they have been in startups for long periods, does this mean they worked at a chain of failed companies?
No no no :) They want(ed) just change :)
I wish you study the Wikipedia page, please. it's a very old company and we have invented really cool stuff.

signed: a proud IBMer

What makes you proud about IBM currently? IBM’s “RA” culture fostering Machiavellian politics, singleminded view of next quarter’s success, and asinine yet draconian bureaucracy are three out of MANY reasons why I’m embarrassed to have had IBM forced into my career and got the hell out as fast as possible each time.
[delayed]
easy.

"claude, get this c++ program working"

these types of moats are all but obliterated.

I have worked for IBM, I do gotta say I was impressed with what they are doing in storage.
Not an IBMer, but IBM MQ and Sterling Connect:Direct are both solid, I use them daily and they work wonderfully, although they can be a real bitch to set up and use in certain scenarios.
A bit of hyperbole by me since db2 was made in 1983. But MQ was created in 1993. Sterling was an acquisition. The majority of products that actually work were acquired. Very little if anything of value since 2005+ that was actually IBM created. (Bob appears fully IBM created)
how long before some hero puts them on github?
Does anyone actually care about what IBM's skill files say?
What do you mean by renting out an agent skill? How can one even rent a Skill? It’s just a markdown file
It could be added to context only on the inference server, with the model heavily instructed not to repeat its contents. Not foolproof, obviously.
You periodically pay a fee that gives you the right to use a licensed markdown file, in the same way you can rent movies or fonts.

When you stop paying but keep using the skill, that’s called pirating.

But if you point your LLM at it and say you want something like this, then it’s all ok. :) If you can’t legally copy it, just launder it through AI.
depending on what the skill does, it can be a whole collection of python/other code files, resource files (i.e., js libs, css/html templates) and markdown files with orchestration instructions
This is the most IBM thing IBM ever IBM'd.
I'm really confused what this is, is it a harness or a new model by IBM? Some reviews on that page suggests that it's a new model
I believe it sends to various models in the background, So largely a harness / framework. Ibm does have its own model which is one of the options for Bob to use.

(Note, I'm an IBMer but nowhere near this so my comment is strictly from my public knowledge/awareness:).

so it routes to actual AI companies and IBM adds additional cost on top. what a business model.
(comment deleted)
The review that "IBM Bob isn't just another autocomplete tool" tells you everything you need to know about this product. lol.
31 years after Microsoft Bob, IBM releases it's own IBM Bob. :)
I thought MS killed that name forever.
(comment deleted)
The companies who are hoarding RAM to produce pointless products should be called out. Deprive working class from computing resources to sell some corporate crap. Get fucked.
If they produce pointless products then they will lose money.
That money will not "trickle down" in either way.
Not to shit on their effort or anything, but it looks like a VS Code fork + an agent panel which is very ... late 2024 energy?
I’m quite certain it’s a fork of Roo Code which is a fork of Cline.

When I saw earlier reviews and videos of it, I couldn’t shake the ick of IBM shamelessly forking a startup’s code. There was even some UI slop where they had some big green pulsing button.

That combined with knowing full well how inefficient the default Roo Code base prompts were, I would not be surprised if there’s some massive token churn.

Oh, so we are yet again in the 'Bob' phase?

Maybe this time it'll work.

(comment deleted)