New
Have AI coding assistants changed how you review code?
More PRs, bigger PRs, new review checklists? Share what changed in your process and quality metrics.
Evaluation guide
How to evaluate AI agent platforms
This guide is for engineering, operations and transformation leads deciding whether to buy an agent platform, build on a model provider's tooling, or extend the automation they already run. It suits teams moving past pilots toward agents that take actions in production systems.
In short
Use this guide to compare AI agent and automation platforms before you commit. Judge them on how they limit what agents can do, how you test and trace behaviour, how they connect to your systems, cost per completed task, data handling and how easily you can change models. Then shortlist three to five companies on Techarda and test them on one real workflow.
What’s changing in AI Agents & Automation
Before you compare anyone, it helps to know what has shifted in the market, because it changes which questions matter.
Agents now take actions
Agents call APIs, update records and trigger workflows. The risk has moved from a wrong answer on screen to a wrong action in a system of record, which changes what you need from security and audit.
Open protocols for tool access
Standards such as the Model Context Protocol make it much easier to connect agents to internal systems. They also make it easier to hand an agent more access than it needs, so permissions deserve early attention.
Category lines are blurring
Workflow and RPA vendors are adding agents, model providers are adding orchestration, and business application vendors are building agents into their own products. You may be comparing three very different kinds of company.
Pricing is moving to usage and outcomes
Seats are giving way to charges per task, per action or per resolved case. That can line up cost with value, but it makes budgets harder to forecast.
Governance expectations are firming up
Frameworks such as Australia's Voluntary AI Safety Standard and, for teams operating in Europe, the EU AI Act are shaping what boards and regulators expect to see: testing, human oversight and records.
What to judge vendors on
Those shifts shape the criteria below. Each one says why it matters and what to ask, so you can take it straight into a vendor call.
| Criterion | Why it matters | What to ask vendors |
|---|---|---|
| 1Permissions and guardrails on actions | An agent with broad access can do broad damage. You need to limit what each agent can touch and decide which actions need a person to approve them. | “How do we restrict which tools and records an agent can use, require human approval for specific actions, and revoke access immediately?” |
| 2Testing before release | Agent behaviour changes with prompts, models and data. Without repeatable tests you find regressions in production. | “How do we build and run test sets for our own workflows, and can we block a release when results fall below a threshold we set?” |
| 3Tracing and audit trail | When something goes wrong you need to see exactly what the agent saw, decided and did. Auditors will ask the same question. | “Can we see every step, tool call and model response for a given run, export it to our own logging, and keep it for our retention period?” |
| 4Integration with your systems | Most of the effort in an agent project goes into connecting systems and handling authentication, not into the model. | “Which of our systems have maintained connectors, who maintains them, and how do you handle acting on behalf of a user compared with a service account?” |
| 5Model choice and portability | Models improve and prices change every few months. Being tied to one model or one platform limits your options at renewal. | “Which models can we use, can we bring our own endpoint, and what would it take to move our agents to another model or platform?” |
| 6Cost per completed task | Token use, retries and tool calls add up in ways a seat price hides. The number that matters is what one successful task costs. | “What does a comparable workflow cost per completed task, including retries, and what spending limits and alerts can we set?” |
| 7Data handling and privacy | Prompts and outputs often carry customer and employee data. You need to know where it goes and whether it trains anything. | “Is our data used to train any model, where are prompts and outputs stored, and how is personal information handled during a run?” |
| 8Failure handling and hand-off | Agents fail partway through multi-step tasks. What happens next decides whether the failure is an inconvenience or a data problem. | “What happens when an agent fails halfway through a task, how are partial changes undone, and how does a person pick up the work?” |
Trade-offs to settle early
No vendor does well on every criterion, and some pull against each other. Decide where you stand on these before the demos start, or the demos will decide for you.
Buy a platform or build on model provider tooling
A platform gives you connectors, governance and a console on day one. Building on a model provider’s tools gives you more control and fewer layers, but your team owns testing, tracing and permissions.
Autonomy or oversight
Every human approval step adds safety and removes speed. Start with approval on anything that changes a system of record, then relax it where the track record supports it.
Tuned for one model or portable across many
Platforms built around one model often get more out of it. Portable platforms let you switch as prices and quality shift, sometimes at the cost of features.
Extend what you have or add a specialist
Your existing workflow or application vendor already has your data and your trust. A specialist may be further ahead, but it adds another supplier, another security review and another contract.
Speed to pilot or readiness for production
The fastest demo is rarely the platform that handles permissions, testing and audit well. Weigh the second more heavily if the agent will act on real systems.
Red flags
As the answers come back, watch for these. One on its own isn’t a deal breaker, but it deserves a follow-up question in writing.
- Demos run only on curated data, and the vendor avoids testing on your workflow.
- No way to see or export step-level traces of what an agent did.
- Agents run under one broad shared service account.
- Success is reported as conversations or interactions rather than completed tasks.
- No clear answer when you ask if your data trains any model.
- Usage pricing with no spending caps or alerts.
Build your shortlist
Pick the workflow first, then use Techarda to find three to five companies worth testing against it.
- Choose one real workflow with a clear definition of done, and list the systems it touches.
- Keep companies that already integrate with those systems. Roadmap connectors do not count.
- Include one vendor you already use for automation or cloud and at least one specialist, so you test build against buy honestly.
- Read recent stories for pricing changes, acquisitions and product retirements. This category moves quickly.
- Take the open practitioner questions below into your vendor calls, then compare side by side.
Companies to consider in AI Agents & Automation
Listed by recent activity on Techarda (what people are reading, following and discussing). The order says nothing about quality, market share or fit for your needs.
Latest AI Agents & Automation stories
Once you have names, recent news is where pricing changes, acquisitions and outages show up first. These are the five newest stories in this category.
AI Agents & Automation
Caller ID in the age of agents
Introducing fleming-1, Sierra’s own model for detecting when the caller on a phone call is another agent.
AI Agents & Automation
Marc Benioff on the AI boom, SaaSpocalypse, and future of Slack
AI Agents & Automation
Agents that can pay: building Restock with Stripe's Link and Managed Deep Agents
Build an agent that pays for real purchases. Restock runs in Slack on Managed Deep Agents and pays with Stripe's Link over the Machine Payments Protocol.
AI Agents & Automation
AI adoption: A practical guide from Zapier
I have two voices in my head when it comes to new tech. One is the wide-eyed optimist who wants to try every tool and click every AI button. The other wants to delete every app on my phone, move to a cabin in the woods, and write by candlelight. There's no doubt AI is exciting. It's already helping teams save time, be more creative, and get through the workday with fewer repetitive headaches. But…
AI Agents & Automation
The best AI newsletters in 2026
As a writer covering technology and AI, I spend plenty of time separating useful information from the hype. A good AI newsletter can help cut through that noise. But the best newsletter depends on what you need. Someone making AI strategy decisions needs different coverage than someone looking for an AI workflow they can try this afternoon. To find the best AI newsletters, I started by…
What practitioners are asking
News tells you what vendors announced. These open questions show what teams are still trying to work out, with the least-answered first.
New
Which security investments are you prioritising for FY27?
Identity, cloud security posture, AI security, SOC automation, resilience? Share your top two and why.
New
How are you evaluating AI features in your ERP/CRM suites?
Vendors are bundling copilots into every suite. How are you deciding what to switch on, and how are you measuring value?
New
How are you approaching AI governance and model risk management?
Inventories, risk tiers, approval workflows, monitoring. What framework did you adopt and what does day-to-day governance look like?
New
Which workflow did you automate with an AI agent first, and did it stick?
We are hearing a lot about agent pilots and far less about agents still running six months later. What was the first use case, what guardrails did you need, and is it still in production?
New
Which skills are you hiring for in FY27 that you were not a year ago?
AI engineering, platform, FinOps, security automation? Share what is on your hiring plan.
Where evidence is thin
A shortlist is only as good as the evidence behind it, and ours is still growing category by category. Here is what’s missing in AI Agents & Automation right now.
20 of 20 companies have no published reviews
If you’ve run one of these in production, a short review helps the next team decide. Reviews are moderated before they appear.
7 questions with no answers yet
An answer from someone who has made the same decision is often more useful than any guide. Share what worked, what didn’t and what you’d check next time.
Answer a AI Agents & Automation questionFrequently asked questions
What should I ask AI Agents & Automation vendors about permissions and guardrails on actions?
An agent with broad access can do broad damage. You need to limit what each agent can touch and decide which actions need a person to approve them. Ask: How do we restrict which tools and records an agent can use, require human approval for specific actions, and revoke access immediately?
What should I ask AI Agents & Automation vendors about tracing and audit trail?
When something goes wrong you need to see exactly what the agent saw, decided and did. Auditors will ask the same question. Ask: Can we see every step, tool call and model response for a given run, export it to our own logging, and keep it for our retention period?
What should I ask AI Agents & Automation vendors about model choice and portability?
Models improve and prices change every few months. Being tied to one model or one platform limits your options at renewal. Ask: Which models can we use, can we bring our own endpoint, and what would it take to move our agents to another model or platform?
What should I ask AI Agents & Automation vendors about cost per completed task?
Token use, retries and tool calls add up in ways a seat price hides. The number that matters is what one successful task costs. Ask: What does a comparable workflow cost per completed task, including retries, and what spending limits and alerts can we set?
This guide is written by Techarda editors. No vendor paid to appear in it or saw it before publication. See our methodology, the trust dashboard or go back to the AI Agents & Automation hub.