Article Content

I am here to-night to warn you, that you have yet a chance of escaping my fate. A chance and hope of my procuring, dear reader.

I know, I'm a good friend.

You will be haunted by three spirits. Without their visits, you cannot hope to shun the path I tread. Expect the first to-morrow, when the bell tolls one.

Expect the second on the next night at the same hour. The third upon the next night when the last stroke of twelve has ceased to vibrate.

What's that? "Couldn't you take 'em all at once, and have it over?"

Eh, go on then.

A photo taken from the audience of Matt Garman's keynote, of a slide featuring three new releases: The Kiro Autonomous Agent, the AWS Security Agent, and the AWS DevOps Agent.

This slide basically sums up your big hitters, as I'm sure you've seen elsewhere. I anticipate that they will completely shift the software engineering landscape over the next few months (they're already doing so). While the environmental impacts of AI don't really fill me with glee, I think we're at the crossroads now — it was always inevitable. It's now not a case of "Should we use AI to augment our software engineering", but a case of, "We're going to have to use agentic AI to stay relevant whether we like it or not, so how can we work around the problems that come with it".

So, this is a little late. But pretty much since re:Invent, I've either been recovering from jet lag (I had almost 24 hours of travel each way 🫠), trying to quell the ringing in my ears, or running around like a town crier, bleating about the Rapture of Kiro.

Or off on my festive jollies; cut me some slack.

First Impressions

This was my first time at re:Invent, actually. First time in Vegas, as well. To digress for a minute — bloody hell, the layouts of those casinos would not be allowed in Europe 😆 Not a fire exit to be seen when you're wandering around the Venetian casino bit. On that note, there was an egregious number of signs and folks wearing "re:Invent" flags around the venues, so to the organisers' credit, the absolute labyrinth that is Vegas was made surprisingly navigable.

As for the event itself... it was good fun. Goes very quick, but it doesn't half knacker you. I think I racked up about 70km of walking over the course of the week — and I used the shuttle bus a fair bit, too. I'm not the most extroverted of folks either, so setting off at 7AM, getting a full day of talks, workshops and such in, and then going to a networking event shortly after... You conk the minute you hit the pillow.

I could yap for a while, so I'll list some of the things that came as a surprise:

  • It doesn't feel like you're there for work. This is something I've seen with other AWS events I've been to as well — they really do nail the art of not making it feel like a chore. No, I'm not just talking about the free beer. Yes, the free beer was cool. It feels very much like, "We know you think AWS is cool, but we also know being talked at for hours on end is pretty dull, even about something you find interesting". There's an atmosphere of genuine excitement — not just the superficial, corporate "Yes, my 9-5 grind is ecstasy, and no, my smile isn't being held up by clothespins!" kind of excitement.
  • Christ on a Las Vegas Boulevard-circuiting bike. The scale of the thing. I wasn't actually staying in one of the re:Invent discounted hotels, like Mandalay Bay, etc. — but the staff sure as hell know about it. Half the folks you walk past have a re:Invent pass on, and the venues genuinely do span the entire strip. It's quite hard to comprehend.
  • The keynotes get... busy. Particularly for Matt Garman and Werner Vogels. I think I queued up about 90 minutes before both, and the queue ahead of me was still pretty hefty.
  • Events (non breakout session ones) also get snapped up quick. I was quite late to the party, so I was mostly stuck with walk-ups. I'd queue about 20-30 minutes prior for most things, and I'd always get in — so it's not a huge show-stopper. You just have to plan your time a lot better.
    • I went to mostly chalk talks and workshops — I'd recommend going to those over the likes of breakout sessions, since they're less interactive and are published online later anyway.
  • The certification lounge is definitely the best retreat away from the noise. As long as you have at least one AWS certification, you're eligible, so make sure you link your certification account and your re:Invent account!

The Takeaways

Though I haven't been to re:Invent before, I'm at least aware that there was an aggressive focus on AI at the last two. But what's a little different this time around is that even your regular cutting-edge (but not agentic) GenAI seems to have been swept to the sidelines a little bit. The focus this year was palpably agentic AI. The heaviest hitter was pointedly Kiro. I mean... they built a whole "haunted house" for it.

A photo taken from inside the House of Kiro, depicting a vacant, dilapidated desk with three monitors. Displayed upon them are Kiro's capabilities; building new apps, building upon existing apps, and scaling apps.

I'm not sure whether this was deliberate or not, but although the House of Kiro wasn't conventionally scary, my interpretation of this desk was quite frightening indeed. The derelict workspace seemed almost to imply a future of redundancy; one where my desk is abandoned and my career as a techie is no more.

That's the pessimistic outlook, but as Werner Vogels' keynote would tell you... yes, I'm correct to be pessimistic.

A photo taken from the audience of Werner Vogels' keynote, of a slide asking the question: 'Will AI take my job', and the answer: 'Maybe'.

...But, as Werner would go on to emphasise, it's only going to take your job if you don't adapt. Which is a sentiment I've been expressing for a while, too.

I'll digress again a little here (this is probably something that should go on the thoughts page, I know) — I've only been in the field for about five years. I haven't personally seen huge optimisations and industry shifts like much more battle-hardened veterans have. But I'd like to think I pay quite a lot of attention to these things anyway... and my conclusion is that everything is always just an abstraction. The plough didn't make the farmer redundant. Mechanised production lines didn't make engineers redundant. Low code didn't make the web developer redundant. People have to change, sure. But history repeats itself. Doesn't seem too threatening to me.

Anyway, you've seen the big three: The Kiro Autonomous Agent, the AWS Security Agent, and the AWS DevOps Agent. As awesome as it is, I think this goes substantially beyond Kiro (which hopefully you've had chance to get accustomed to yourself — AWS has made it their de facto IDE, and I've as good as done the same).

Spec-driven development enables Kiro to address the fundamental problem with vibe-coding, which is:

Humans suck at giving instructions, and AI is great at following instructions.

But Kiro still feels supplementary to the developer workflow in just the same way as GitHub Copilot or Amazon Q Developer do (larger-scale, yes, but fundamentally similar). The three ghosts of re:Invent represent a shift away from "AI writes the code, developer handles the logistics", to "AI writes the code and handles the logistics, and the developer is an overseer".

This kind of plays into something that Werner hinted at, but didn't explicitly state with his Renaissance Developer theory — that developers will naturally shift towards being more architectural.

The obvious eyebrow-raiser here is how such freely-acting agents (I'm particularly looking at the DevOps Agent here) will behave with live, sensitive, fragile and delicate systems, and how well they'll integrate with distributed systems with a diverse tech stack, data sources, communication platforms, and so on. It's difficult to digitise/quantify the context a human engineer is privy to when they work with systems — that's why hand-overs are hard and bus factor exists.

My conclusion being... I'm nervous. Not about my livelihood; I'll figure that out. I'm nervous because I'm a control freak, and with great power comes great responsibility... but AI doesn't have any responsibility, you do. How do we toe the line between AI having little enough power that it can't be culpable, without humans becoming the bottleneck? I don't know.

But the harsh reality is, we have to start using these tools, whether we like it or not. There is no such thing as an artificial stalemate. It's the reason mutually assured destruction has become the reluctant answer to international peace, and it's the reason we can't sweep agentic AI under the rug because we don't trust it.

We can't brush it aside because it writes code that doesn't understand nuance. We come up with solutions, like spec-driven development.

We can't brush it aside because we're afraid of what agentic AI might do during an outage. We will to adapt.

Because whether we like it or not, someone will use it, someone will do the job quicker than you, recover better than you, and their business will live longer than yours.

A photo taken from the audience of Werner Vogels' keynote, of a slide stating that 'The work is yours, not the tool's.

The Hidden Gems

I probably didn't say anything in the above section that's about to give anyone heart palpitations. There were a couple of somewhat more nuanced things I think are worth highlighting, though...:

  • The North Virginia AWS outage from back in October seems to really still be weighing on people's minds. I attended a session on the European Sovereign Cloud*, and I expected ~90% of people there to be from Europe. I think it was more like 60%. Still a much greater proportion than in other sessions, but I was surprised to see ~30% of people seemed to be from North America (in fairness, I'm basing this on folks who spoke up in the session). The thing on a lot of the non-Europeans' minds seemed to be sovereignty solutions in general — Outposts, Dedicated Local Zones (DLZs), etc. I think there were more questions here than in any other session I attended, and the most common reason appeared to be concern around the October incident.
    • Personally... I think this a slightly knee-jerk reaction — it's always been understood that North Virginia is a sort of central region, and I think you'll have much better luck trusting in AWS' ability to maintain things as they do pretty damn well... several nines of the time.
    • *Also, I think there's a bit of a distinction to be made here... The European Sovereign Cloud appears to be more of an EU Sovereign Cloud. Which is a particularly important thing to note for folks like me, because the UK is no longer in the EU. This wouldn't address sovereignty concerns of folks in the UK, Switzerland, Norway, and a few other countries on the European continent, but not in the European Union.
    • There was also the announcement of AI factories at Matt Garman's keynote, which is basically sovereign AWS AI compute. Really seems to be kind of a trend here...
  • I managed to squeeze into one of those new 500-level (Distinguished) sessions. There weren't many of these sessions, and they got quite busy. Mine was positively brain-entangling. By which I mean, I attended a session on quantum networking. I don't... exactly understand quantum, but I get the basic principles. We covered the impracticality of current-day quantum signal repeaters, and the limitations that places on transmission distance and signal loss. The curious thing being that intra-region (i.e., between AZs) quantum networking is theoretically feasible today, given data centres within an AWS region happen to fall just under the typical transmission distance limit. Inter-region networking is where you either want repeaters, or satellites — the latter of which exist, but of course come with their own problems (like weather — clouds! Oh, the irony). It was quite cool, but I imagine for someone genuinely well-versed in quantum, it'd be child's play.

My personal favourite announcement: S3 Vectors. Eh, okay, it was already in Preview before re:Invent, sure. But it became GA as of re:Invent. Now, I'll hold my hand up and say I've never used Pinecone. Or OpenSearch (well, I've used OpenSearch, but not for vectors). There's partially a reason for that — I don't do anything requiring RAG, and they can get expensive, they're a faff to maintain — cue every pain point with managing a database, basically. Not worth it, unless such a requirement comes in from a client. But, I got chance to play around with the Nova Multimodal Embeddings, and Nova 2 Omni, in a workshop. So I thought I'd come home and put that to use.

I got Kiro to write me out some scripts (the scripts are aggressively vibed, no specs — I know, naughty naughty. They are as a result hanging together by duct tape, lol) — one to convert a bunch of files I fed to it into embeddings, and then drop them into an S3 Vector bucket. Then another script that uses Nova 2 Omni to create a query embedding against those Vectors.

Lo and behold, for less than $0.05 (and I invoked and tweaked these scripts a lot), a bumbling (yes, honestly, I had no idea what I was doing) developer as an overseer, and 0 ongoing maintenance, I can use Nova 2 Omni to query a bunch of PDFs, containing the text and photos I collated from my re:Invent escapades.

So after all my waffling, what does this little contraption have to say about re:Invent 2025...?

🚀 Nova 2 Multimodal RAG System is ready for queries!
📚 Using existing vectors in S3 Vector store
💡 Ask questions about your re:Invent 2025 documents!
⚡ Type 'quit' to exit.

Your question: As a very brief, maybe one or two sentence summary, how would you summarise AWS re:Invent 2025? Any big annoucements or themes you'd call attention to?

🔍 Searching for relevant information...
Processing query: As a very brief, maybe one or two sentence summary, how would you summarise AWS re:Invent 2025? Any big annoucements or themes you'd call attention to?
Found 5 relevant documents
Found 5 relevant documents
Found 10 relevant documents

============================================================
📝 ANSWER:
============================================================
Based on the provided context from AWS re:Invent 2025 documents, a brief summary of the event would highlight the focus on advanced cloud solutions, particularly hybrid cloud strategies for digital sovereignty, and the integration of agentic AI for autonomous networks. Notable announcements include enhancements in S3 functionalities, such as automatic replication and faster vector indexes, and the introduction of EMR Serverless, which simplifies storage provisioning. The theme of leveraging AI to drive autonomous and efficient cloud operations is prominent.

**Documents referenced:**
- HMC316 - Address digital sovereignty with hybrid cloud solutions.pdf
- IND302-R1 - Agentic AI for Autonomous Networks - AgentCore Design Patterns in Action.pdf
- KEY005 - A Special Closing Keynote with Dr. Werner Vogels.pdf
- KEY001 - Opening Keynote with Matt Garman.pdf

Not bad. Although it's very obviously biased based upon the particular sessions I attended 😆

And that's all, folks. Thanks for reading. Here's to a fruitful 2026 with no developer redundancy. 😉