Pluspunten
There are really friendly people at CognitiveScale, and in some teams some surprising diversity. This diversity could easily diminish if taken for granted, but in my experience is notably better than our industry's average. It's delightful. Lunch is provided daily, and it's nice enough that lots of people actually come together in a way that feels nurturing, communal, and not coerced. I've enjoyed so many of the conversations we've had over lunch. The marketing and sales teams are also remarkably effective. Though I feel published marketing/sales copy have at times slipped uncomfortably towards untruths, and customers have been hooked with promises I'd expect no trustworthy ML expert to sign up for, it's hard for me to blame the marketing and sales teams directly. I'm convinced they're getting their information from senior management who assures them that these statements and promises are grounded, true, and possible. The design team also has talented and caring people. And the project management team works diligently and compassionately to make things work respectably despite challenges. There is also a team of highly trained, experienced, and respectable ML experts delivering honest one-off solutions to customers. Some of them also pursue research that may develop into interesting products. Unfortunately, I believe the margins on these consulting ML engagements are too low to satisfy the investors' ambitions. It seems to me CognitiveScale has to sell a more expensive engineering product or platform, which leads to the cons of working there. My review of CognitiveScale in the "Cons" section below is scathing, I know. But my critique targets mostly two aspects specifically: the leadership teams and the engineering. If you can find employment at CognitiveScale that's distanced from these, then you may find fulfilling employment despite deep systemic problems. For instance, maybe you can do good design work on the design team. Or maybe you can do solid ML work with the ML team.
Minpunten
You may think that having tons of money and being in a hot market would be an asset. Unfortunately, as far as I can see at CognitiveScale this has spoiled the executive staff and investors to a degree that has depressed the morale of those employees interested in professionally ethical or quality work. I'm inclined to characterize the executive staff as a slew of yes-men (women seem notably pushed out of this inner circle of trust). Many appear promoted despite clear examples of gross incompetence and ignorance. From my vantage point, loyalty and even nepotism trumps all. I used to be shocked at how little almost every executive and manager knows about AI for a company selling AI products and solutions. As time passed, I grew numb to the ignorance and was less shocked. Leaders with only a lackluster engineering background boldly make declarations like "I've done ML"—even among published researchers—on the basis that they wrote a something like a regression a few times or packaged an algorithm someone else wrote. Expertise for these people is no more than a baseless bullying assertion. The few leaders that have more expertise either seem unable to notably improve the company, or have left. In my estimation, the most senior engineering staff is woefully technically unskilled, spending more time posturing than improving their technical skills. In my opinion, they couldn't tell you the fundamental difference between Kafka or RabbitMQ and why you might choose one over the other. Also in my opinion, they couldn't explain why putting small files on HDFS was a bad idea. These are the kinds of people who find creative and dangerous ways to use MongoDB. As far as I can tell, they are "big idea" people. These sort of people take pleasure in stunning an audience with sprawling UML diagrams of these ill-advised big ideas. They hurl these architectural plans at unprepared engineers who aren't skilled or senior enough to push back, and instead just take a stab at implementing these byzantine designs. All of this appears to me a replica of the incompetent technatrocities I'd expect in the worst parts of IBM, but in miniature and without the massive funding of IBM to keep it afloat for as long. For the most part, CognitiveScale seems to have installed the same team that's "worked" for them in the past, including senior staff from Webify and IBM Watson. These prior initiatives may have accrued investors millions, but in my eyes they were ultimately examples of deeply flawed technologies, executed hastily and driven more by marketing, smoke, and mirrors. It's easy to Google for "IBM Watson fraudulent" and "IBM Watson joke" and find some incisive commentary. I'd characterize CognitiveScale as literally a worse sequel with roughly the same plot and actors, but a markedly smaller budget. I've detected a pattern of CognitiveScale taking industry buzzwords like "knowledge graph," "blockchain," or "ethics in AI" and making visually stimulating demos, too often with no connection to a worthwhile AI project I could see. These demos seem to raise investment money or hook a customer. Then the company seems to force the demo into a production mode, find out it isn't a good idea through the harsh forces of reality, abandon the idea, and start over with a demo based on another hyped term. All-hands meetings often remind me of the tale "The Emperor's New Clothes." As I see it, externally projected marketing is turned inwards, amplifying as it resonates within the walls of the office, deafening us with absurd sloganeering. I've seen executives roll out what looked like UI-only demos, animated like the ridiculous flashy systems one would see on a modern crime show -- the AI equivalent of an HTML blink-tag. With no apparent AI system backing them, the high level executives talked about these baubles as though they would actually become part of the product and drive business. I'd just sit back in my chair dumbfounded, and watch over time as my expectations of these travesties unfolded into reality and predictably went nowhere. CognitiveScale's revenue streams seem to be fueled by enterprise executives with deep pockets, but little clue of what they are buying or how to measure the value of it. As far as I can see, CognitiveScale simply takes as much money as they can from ignorant clients for as long as they'll let them. The one-off ML solutions are at least of some value. CognitiveScale's consulting work does indeed connect clients with highly skilled ML experts motivated and capable of delivering value. But when CognitiveScale upsells the customer to the platform, by my ethical compass, they are leading the client down an exploitative path. Like kids shoplifting from a giant store, they can pretend that their take is a negligible portion of all the money flowing through the market; they can pretend that no one gets hurt. But these tiny thieves damage society in aggregate. Collectively companies like CognitiveScale drive prices higher for the common consumer by charging high prices for what in my opinion is a low- or negative-value product. We can claim this is part of the process of innovation. But I would opine that anyone from a deservedly respected company in the AI space could see it dangerously close to something far worse… fraud. Some people working currently and previously employed by CognitiveScale have read John Carreyrou's book Bad Blood about the scandalous story of Theranos and the massive deceptions of its leader Elizabeth Holmes. It's hard to read this book and not see a myriad of parallels with CognitiveScale. A lot of us who work or have worked there are looking forward to seeing the film adaptation when it comes out. With engineering like this, it's no surprise that customer churn is much higher than I'm comfortable with. CognitiveScale gets stuck in a cycle of short-term firefighting, and doesn't plan for the technical requirements and execution of long-term support that would address client churn. I see no way this process can scale. This is where the platform is supposed to help, but to date it's not inspired my confidence. From a distance, I may admit the platform vaguely resembles more well-known AI platforms, but closer inspection quickly reveals to me ominous fissures. The platform is supposed to offer a palette of pieces that compose to make a client's solution. But when various experts assert that the pieces have no hope of actually composing usefully, the leadership and engineering staff dismiss the problem. They seem to just want a large collection of AI-related pieces, assuming that clients and investors won't figure out the deeper problem until after funds have been released. The platform feels like a bad student group project for a class no one wanted to take in the first place, cobbled together at the last minute, doesn't do what we'd expect, and loaded with padding to distract from the lack of substance. These are the "D" students relying on the teacher to pass them anyway. I'm astounded at how much work has been thrown away with respect to the platform over the last five years. They just seem to reset every year or so with very little code, learning, or introspection to carry over. The latest iteration of the platform strikes me as an unimpressive jumble of services. Much of the visual user interface frightens me with complexity, and I can barely figure out how to do data processing with it. It definitely doesn't make clear to me how any of it helps a real-world AI workflow, and I'm convinced that's because it really doesn't. There's always a piece of the platform that people swear is on the verge of being what we really want. But after being let down by these same people so many times, I've lost faith they'll deliver on that possibility. I would guess people on the team that made it know the platform has huge robustness, performance, design, and quality problems. I would not at all be surprised if they said comments like, "let it fail hard with the customers, and we'll learn where the 'kinks' are." Another phrase I'd expect is "maybe it is bad, but it pays the bills." And by my measure, many people don't have the expertise to know how bad it really is or how to correct it. I've gotten a sense of the range of pay from talking to employees at CognitiveScale (without disclosing exact numbers). Some jobs do have competitive salaries. But management seems to treat many of their common technical staff like cheap and fungible resources. This is the kind of management whose personal philosophies lead to positions like, "Stop being so picky. This is Austin. We've got tons of technical people. Almost anyone will do." But they also say contradictory (and laughably hyperbolic) things like, "We have the best of the best." If I were making an AI platform, I wouldn't even consider such a feckless approach to talent acquisition. As one might expect, the attrition rate for CognitiveScale's most talented people is far too high for my comfort, especially recently. I feel these high-value employees held on for two reasons: they thought they could make a difference, and they loved working with other talented people. But through the course of witnessing the executive culture and engineering decisions, I feel these people became so bitter they didn't recognize themselves. On top of this, I sensed they feared atrophying skills. I'm sure many of them voiced their grievances in many different modalities and forums. But it's demoralizing when leadership with insufficient training or experience casts their own terrible short-term ideas as equivalent in value to the long-term ideas conceived by the company's most talented and experienced staff. Ultimately, you can wake someone who's asleep, but you can't wake someone who's pretending to sleep. I understand why so many of the good ones left. In my opinion, most who thrive at CognitiveScale are delusional, or amoral, or uncommitted to a product of any appreciable value. They seem to only want the perception of value for an inevitable acquisition. Actual value for the consumer may be nice to have, but not essential. Note, to avoid defamation I've written everything in terms of my opinion, with no real verifiable facts, true or false. This unfortunately casts me as cowardly and arrogant. I really wish we lived in a world where as I could say more explicitly what I've witnessed and heard from reliable sources. I've worked in a few places, and all places have their problems. But this place stands out to me as an absurd extreme, like an episode of HBO's Silicon Valley. Unfortunately, I can't just laugh it off because I also find it unethical. Finally, some of the positive reviews here on Glassdoor seem to me extremely suspicious. They read like the work of an executive or other stakeholder trying to cheat the system.