Two in three respondents say their company is considering replacing a vendor or tool with an AI agent. In our September survey of 1,030 people who build, buy, or influence AI systems, 26% said they are weighing replacing many vendors or tools, and 41% said one or two. Another 17% plan to extend the tools they already have. Only 11% have no plans.
If you sell software to these companies, that is your next renewal conversation. The same survey shows what buyers would switch for, what slows them down, and how much they have already built on their own. Together, those answers point to a narrower and more specific risk than the usual headline about AI agents replacing vendors.
They have already started building
81% of respondents have replaced at least one vendor or tool with in-house tooling, and 32% have replaced six or more. The question did not ask whether the replacement was an agent, so we read it as a measure of willingness to build. That willingness is high.
Engineering capacity sits low on the list of what holds them back. Asked what is preventing agentic AI implementation today, 24% named engineering bandwidth, fifth on the list. Security approvals (42%) and integration complexity (40%) lead it, and 23% say nothing is blocking them at all.
What they would switch for
38% say they are very likely to switch tools or vendors for stronger agentic AI capabilities, down from 46% in December. Another 55% say possibly, so the share open to switching has held (93.6% then, 93.0% now) while the urgency has cooled.
We asked the 958 respondents open to switching what they would switch for, with one answer each.
What they would switch for
Share (n=958)
Reliability
23%
Security and compliance
21%
Better integrations
19%
Lower cost
13%
More autonomy
10%
Better UX
8%
Vendor trust and support
6%
The question asked about agentic AI capabilities, and more autonomy came fifth. Buyers will move for an agent. The agent they want works every time, passes a security review, and connects to the rest of their stack. Those three answers add up to 63% of the reasons. Price and autonomy together add up to 23%.
Platform selection tells the same story. Asked what matters most when choosing an agentic AI platform or API, with up to four picks, 65% chose reliability and 61% chose security controls. Performance (50%), ease of integration (46%), and cost (41%) follow.
Why reliability comes first
The people answering have watched agents fail. 54% had an AI agent do something wrong at their organization in the past 90 days, most often by triggering a workflow the team had to reverse (31%). A buyer who has cleaned up after one agent judges the next on whether it breaks.
Where a vendor gets replaced
Put those findings side by side and the risk gets specific. A product is exposed when an in-house team can match its reliability, own the security story, and wire it into their systems without help. 81% of respondents have shown they will build when the case is there. A product keeps its place when it is the more reliable option, ships the controls security review asks for, and costs less effort to integrate than to rebuild.
The 17% who plan to extend existing tools deserve attention too. They want to build agents on top of what they already pay for. For a vendor, that means APIs, webhooks, and permissions an internal agent can work with.
Four things to check before your next renewal
Report reliability as a number. Publish your API success rate, your incident history, and how failures get handled. Reliability is the top reason to switch (23%) and the top platform criterion (65%).
Give security review a packet before it asks. Security approvals are the top blocker at 42%. Document permissions, data handling, and what your product logs about the actions it takes, including actions taken by agents.
Make your product something an internal agent can call. If a customer’s team can build an agent on your API, your product becomes part of their build. That is the route the 17% who plan to extend existing tools are already on.
Lead less with price and autonomy. Together they account for 23% of the reasons to switch. A discount answers a question most of these buyers are not leading with.
The same test applies to us
Nylas sells infrastructure that product teams build on. Teams building agents that read and send email or manage calendars run into the integration complexity 40% of respondents named: separate Google and Microsoft APIs, OAuth edge cases, and sync behavior that changes by provider and surfaces at scale.
Nylas normalizes email, calendar, and contacts across 250+ providers behind one API, with a 99.99% API success rate, measured as successful calls over total calls. The platform is SOC 2 Type II compliant, and Agent Accounts give each agent a mailbox and calendar of its own. The bar we hold ourselves to is the table above: more reliable, easier to secure, and less work to integrate than building it in-house.
58% of teams with customer-facing AI agents let them send as an employee at least some of the time. New Nylas survey data on agent identity, mistakes, and audit.