Happy Friday! I wanted to share some of my initial thoughts & observations from Dreamforce this week.
Salesforce vibe: broadly, muted positivity. I didn't see the outright confusion of some prior years (eg, 'what is Agentforce??') or the excessive corporate kool-aid of some prior years (all relative... it's still Dreamforce). There is a sense of an "identity crisis" - 5 years from now, will Salesforce just be the database with a business process layer on top? - but the Claudeforce announcement was reassuring. Although the future feels unknown, everyone I talked to is committed to Salesforce, personally and professionally.
Customers: to me, a reminder of the bubble I live in. I talked to customers who ranged from currently implementing, to just inherited a very messy org, to has been the admin for 15 years. None were using AI. One was waiting for his board to approve an AI policy, another felt like he had too much on his plate to experiment right now. (I’m not talking about Agentforce - this is any AI.) I’m certain this is not representative, but it is a good reminder of how "AI-pilled" I've become, to use Sri's term. (Only a small single digits percentage of Salesforce customers are using Agentforce, too.)
Partners: I talked to about 50 SIs this week (whew!).
What they have in common: everyone is frustrated with Salesforce. It feels like Salesforce is cutting back on partner support, while also asking them to do more. The naming changes are causing real problems: I talked to one SI who’s in the middle of building Agentforce agents for an enterprise customer. After Salesforce announced they were removing Agentforce from their product names and announced Claudeforce, the customer came back to the partner and said: should we still build these Agentforce agents - or should we build them with Claude?
Packaging changes are also coming too fast and too furious - partners want to be customers’ trustworthy guides, but they don’t have the most up to date information (in fairness, AEs often don’t know either…). (These challenges are being heard by Salesforce leadership - I thought Grant Terrell acknowledged it well during the Nonprofit Partner session on Thursday.)
Partners are feeling pressure to reduce the cost of implementations because of AI efficiencies. One SI I talked to said he's now coming in at half of what he would have charged a few years ago; another said the pricing pressure is not actually from clients, it's from Salesforce AEs. I’m also hearing huge variance in timeline impact - from shaving off a few weeks to projects that would have taken 3 years now taking 4 months (more on that below).
Of course the thing I was most interested in was how SIs are using AI for implementations. The spectrum is wide; I’ve tried to group them into a few categories:
E2E products: some SIs have built their own proprietary AI products that they're using to do their implementations, ranging from an internal-only agentic fabric layer to a user-facing web app product. These E2E tools assist in every phase: discovery, build, data migration, testing, deployment, go-live, documentation. This is where I consistently heard the most realistic implementation time savings (~ 25%ish), where SIs are doing fixed-bid work, and where there was substance when I prodded a little - it seems like they’re being used successfully in the wild. Possible caveats: I heard from some other SIs that there isn’t a lot of room for nuance with this approach.
Partial agentic automation with bespoke tools: these SIs are using Cursor or Swantide or Agentforce Vibes to accelerate build, testing automation tools like Selenium and Zephyr to make testing faster and more comprehensive, and a patchwork of L&D tools for documentation and training. One person told me they’re spinning up ~60% of the build based off the RFP, which helps them win deals. These SIs tended to be US-based with deep industry expertise.
Building their own: Some SIs are building Claude Skills for everything - from writing requirements to doing first-pass data mappings. One SI told me that their process today is unrecognizable compared to what the team was doing 6 months ago, because they’re automating so much.
SIs are made up of builders who love technology; many of the larger SIs I talked to have AI eval teams that are responsible for strategizing, owning, and in many cases, building AI tooling for internal teams. I think AI build v buy has come to a peaceful end for most businesses - but SIs are different, and there is still a strong case for build. Historically, services companies haven’t been great at also being product companies, but internal tooling may be different.
With that said, when I pushed a little on how it’s changed timeline or pricing, I got the sense with some SIs (not all!) that the agent suite they had built was either not being used in production, or was overclaimed.
Not using AI in any meaningful way: some SIs are prevented from using AI at all by their clients, some can only use their client's AI (undoubtedly Copilot), and some prefer not to use AI because they think the quality isn't there. One SI I talked to uses AI to coach her team on writing better user stories, but the team still has to write them by hand initially. That’s slightly different than what I heard a year ago, when some people didn’t use AI in any way.
Looking ahead: There's a big bet right now in venture on AI Native Services businesses - in Salesforce parlance, SIs that are built with AI from the ground up. I've talked to a few new SIs working in other areas (SAP, Hubspot) that are going into this business, and obviously some Salesforce SIs are changing in this direction as well. What's been interesting to me, though, is that we're seeing more products pop up than SIs: products SIs can buy, or enterprises can buy, to automate a lot of the end to end (eg June, Laminar, Auctor). Still thinking on this over the weekend - more to come here soon.
Would love to hear your thoughts - does this align with what you're seeing and hearing? Email me back or grab a time here.
Have a great weekend!
- Ali